An ontology-based approach for supply-chain quality control: From a principal

Abstract

The efficacy of the principal–agent contract in supply-chain quality control depends not only on contract parameters but also such noncontract parameters as cost of a high-quality effort and the diagnostic error of the inspection policy. The noncontract parameters usually fluctuate and are unobservable during contract execution, which may hinder suppliers’ high-quality effort, or, in other words, result in a lower efficacy for the contract. This article proposes an ontology-based approach to facilitating a principal–agent contract by monitoring the contract’s loss of efficacy. The approach consists of ontology-based models and data-centric algorithms. The ontology-based models not only formally represent concepts and relations between concepts involved in predicting whether a contract is efficient, but also organise multichannel data such as news, marketplace reports and industry databases containing information of factors impacting the unobservable noncontract parameters’ fluctuations. Based on the ontology-based models and multichannel data, the data-centric algorithms are developed to predict whether a contract will lose efficacy. We evaluate our approach through case study, simulation and comparison against related approaches to supply-chain quality control. The case study proves that our approach is appropriate. In the simulation evaluation, a combination of our approach and principal–agent contract is more efficient than just a principal–agent contract. The comparison results against related approaches show that our approach is a novel, inexpensive and directly applicable tool for reducing both asymmetric information and moral hazard in supply-chain quality control.

Keywords

Data-centric framework ontologies principal–agent contract quality control supply chain

1. Introduction

In modern economies, many loosely organised manufacturing firms distributed in different places or even different countries cooperate in product manufacturing. Such product manufacturing forms the supply chain, and each organisation is a supplier of the product [1]. In most cases, the supply chain for a product is complex and too long for downstream suppliers to observe and track upstream ones’ product-quality effort; therefore, product quality control forms a challenging problem against the supply-chain background [2]. Many quality scandals have occurred worldwide in recent years. For example, 2008 melamine scandal in China [3], the Germany horsemeat scandal in 2013 [4] and 2014 expired chicken at McDonald’s located in China, Japan and Hong Kong have had extremely negative global effects [5]. The reasons for these quality incidents can be concluded as follows. First, the product’s downstream suppliers have less information than the upstream suppliers about the latter’s quality effort and cannot distinguish low-quality products even using inspection, the asymmetric-information problem. Second, under asymmetric information, in pursuit of profit, the upstream suppliers may experience a moral hazard. That is, they may supply downstream manufacturing organisations with low-quality products while pretending they are of high quality. Thus, mitigating asymmetric information and the moral hazards induced by asymmetric information is crucial for supply-chain quality control.

The most popular approach to eliminating moral hazards is to adopt traceability systems [6] designed to add value to quality control by offering communication links to identify, verify and isolate sources of noncompliance with agreed standards and customer expectations. They motivate suppliers to make high-quality effort by locating the origin of quality disasters without reducing asymmetric information [7]. However, three defects reduce the capacity of traceability systems to ensure high-level quality control in the supply chain. This is why the EU’s fresh-food supply-chain traceability requirements [8] failed to avert Germany’s horsemeat quality scandal in 2014.

The first defect is that traceability systems are far from able to accurately determine quality-disaster liability [9]. The second is that traceability systems record where, who and when the products are processed along the supply chain. However, information of upstream suppliers’ quality effort cannot be captured, although it is vital for quality control. For example, information regarding the feeding of cows is very important for quality control in the milk supply chain, but it is not commonly shared with traceability systems as the feeding information is considered a core business secret. Third, traceability systems are expensive to implement [10]. These defects decrease the capacity of traceability systems to motivate suppliers to produce high-quality products. Thus, traceability systems have their own limitations in mitigating moral problems [6].

Principal–agent contracts are complementary to traceability systems, legally protected and inexpensive [11]. In a principal–agent contract, the downstream manufacturer is the principal and the upstream supplier is the agent [12 –14]. The downstream supplier designs the contract parameters to give the upstream supplier incentive-compatible conditions and induce the upstream supplier to exert a high-quality production effort [11]. Until now, most research related to principal–agent contracts for supply-chain quality control have focused on the effect of contract parameters, including offer price, rewards, penalty, sampling size and acceptance number, on upstream suppliers’ quality effort.

Although principal–agent contracts have proved helpful for supply-chain quality control, there is a major limitation to this approach. We illustrate the limitation with the conclusion from Starbird’s [15] work: the upstream supplier is willing to exert a high-quality effort only when the value of the offer price and penalty is greater than or equal to the difference between the cost of the high-quality effort and the low-quality effort divided by the difference between the probability of a high-quality product passing inspection and that of a low-quality product passing inspection. If the production cost of the high-quality product increases, that calculation may not favour the principal, leaving the incentive-compatible conditions for the upstream supplier unsatisfied. In turn, the upstream supplier becomes unwilling to exert high-quality effort.

This phenomenon is what the contract losing efficacy means in this article. Downstream suppliers may not notice the lost efficacy for the following reasons. First, whether the incentive-compatible conditions for the upstream supplier are established and whether the upstream supplier is willing to exert a high-quality effort depends not only on the contract parameters but also on other parameters not included in the contract, called noncontract parameters in this study. Noncontract parameters include the cost of the high-quality effort, the inspection tool’s diagnostic error and many other factors affecting the transaction. The fluctuations of noncontract parameters are usually unobservable, yet they are determined by multiple factors. Information on these factors is often contained in data from such channels as news, marketplace reports and industry databases. For example, the cost of a high-quality effort is determined by labour cost, materials cost, technology cost and other factors, which can be extracted from news, reports and other sources. This multichannel data can be captured to estimate these noncontract parameters. In turn, by combining this information with the contract parameters, the principal can estimate the contract efficacy. However, this analysis is complicated for as a manual project. Therefore, it is difficult for managers to notice and capture the fluctuations of noncontract parameters without a facilitating tool. For example, weather conditions impact the cost of safely feeding cattle, which, in turn, affects the dairy farmer’s willingness to produce safe raw milk. This change impacts contract efficacy in the milk supply chain. This weather impact on contract efficacy during contract execution is difficult to measure, but it is not insignificant. Second, when the downstream supplier has many upstream suppliers, especially if they are geographically distributed or when the supply-chain network is complex, it is especially difficult for them to manually monitor all the necessary upstream suppliers’ contracts for lost efficacy.

To address these two drawbacks, we propose an ontology-based approach to the principal–agent perspective in this article to monitor whether a contract has lost efficacy during execution and to notify managers so they can make timely changes to contract parameters to maintain contract efficacy during execution. This approach consists of ontology-based models and data-centric algorithms. The ontology-based models represent concepts and relations between these concepts involved in monitoring contract efficacy from the principal–agent perspective, including concepts related to principal–agent contract parameters, noncontract parameters and information on factors impacting noncontract parameters. The data-centric algorithms are developed to identify suppliers who are not willing to exert high-quality effort, who thus produce quality-risky product. In this approach, data containing information of factors that impact noncontract parameters is first collected and mapped to the ontology-based models. Then, noncontract parameter fluctuations can be derived from the ontology-based models. Based on the ontology-based models and the noncontract parameters fluctuations, the data-centric algorithms predict suppliers unwilling to exert high-quality effort, quality risky product and losing efficacy contract to notify a timely change of contract parameters.

This research offers five contributions to the literature. First, the ontology-based approach proposed in this article mitigates the asymmetric-information problem, in turn reducing the moral-hazard problem in the supply chain, as it can reveal information related to upstream suppliers’ (agents’) willingness to exert a high-quality effort for downstream suppliers (principals) to help monitor a contract’s loss of efficacy. Second, as an information technology-based tool, this approach can augment the traditional principal–agent contract: Supply-chain managers can use it directly to adjust contract parameters to ensure a contract’s continued efficacy during execution. Third, the approach’s models are formal representation of concepts and relations regarding principal–agent contract efficacy. Fourth, multichannel information on factors impacting noncontract parameters can be collected, organised and mapped in the ontology models. Finally, evaluations of case study, simulation and comparison against other approaches demonstrate that our approach is appropriate, efficient in ensuring contract efficacy and a novel tool to reduce asymmetric information and moral-hazard problems.

The article is organised as follows. Section 2 presents the related literature. Section 3 illustrates the background of principal–agent mathematical models and data-centric frameworks. Section 4 illustrated the ontology-based approach including the ontology-based models and data-centric algorithms. Section 5 evaluates the case study, simulation and comparisons against other approaches. Section 6 concludes.

2. Related works

The ontology-based approach is developed to facilitate principal–agent contracting by mitigating asymmetric information and moral hazards in the context of supply-chain quality control. Section 2.1 quickly reviews the broad literature of supply-chain quality control. Section 2.2 reviews the literature related to asymmetric information and moral hazard in the context of supply-chain quality control.

2.1. Supply-chain quality control

Issues related to supply-chain quality control are addressed by two streams of research, supply-chain quality management and supply-chain safety and security management. Supply-chain quality management usually refers to the formal coordination and integration of business processes involving all partner organisations in the supply channel to measure, analyse and continually improve products, services and processes and to create value and achieve satisfaction of intermediate and final customers in the marketplace.

The main target of supply-chain quality management is customer satisfaction, and improving product quality is an essential means to reach that goal [16]. Four themes of research pertain to supply-chain quality management: supply-chain communication and partnership activities, process integration and the management of supply-chain linkages, management and leadership, and best practices. The main theses of supply-chain communication and partnership activities involve organisations working closely together and nurturing relationships with other members of the supply channel to share goals and coordinate activities [17 –19]. Process integration and the management of supply-chain linkages refer to extending firm boundaries and transitioning processes to a system orientation encompassing customers and suppliers [20]. Management and leadership influence relationships and operations with supply-chain partners, not only guiding individual company efforts but also encouraging participation and cultivating quality management measurement among all channel members [19]. Best practice involves activities that promote quality in the supply chain such as supplier relations, customer relations, total quality management, team building, ISO 9001 adoption and Six Sigma [16,21 –24].

The topic of supply-chain product safety and security originates from supply-chain risk management, which is magnified in scope and scale in today’s global economy and has received increased academic attention in recent decades [9,25]. Product safety refers to the reduction in the probability that use of a product will result in illness, injury, death or negative consequences to people, property or equipment. Product security problems include deliberate substitution of materials/components, contamination or adulteration of a product, or misrepresenting a counterfeit product as authentic through counterfeit labelling, packaging or instructions [9]. Several tools address issues of product safety and security in supply chains: supplier relations [26], total quality management [27], ISO 9001 and 22,000 [28], and HACCP. Continuous improvement is a foundation to the 3R approach (readiness, responsiveness and recovery) to managing product safety and careful handling recalls is also encouraged [29]. Most of the above approaches are operation-management tactics or principles for managers to improve supply-chain quality control. Managers design their supply-chain activities in alignment with these tactics and principles using of their management expertise. These methods are less effective in the presence of moral-hazard problems induced by asymmetric information or when intentionally low-quality effort corrupts the supply chain [15,29].

2.2. Moral hazard in supply-chain quality control

In economics, principal–agent models are often used to represent economic relationships that exhibit moral hazard. In these models, the principal’s payoff depends on the agent’s behaviour, but the principal cannot observe the agent’s behaviour. In other words, asymmetric information exists between the principal and agent. The agent decides whether to shirk or to exert less than the effort required to maximise the principal’s objective. The moral hazard arises because the agent prefers to exert less effort, but the principal wants the agent to work harder. Moral-hazard problems are common in different types of R&D licencing agreements, economic or political federations, insurance, labour contracting, the delegation of decision-making responsibility and supply chains [15,30]. Principal–agent theory in supply-chain quality control views the buyer as a principal and the supplier as an agent. In the context of supply-chain quality control, the moral-hazard problem occurs because product-quality information is asymmetric because buyers (principals) have less information about the supplier’s (agent’s) quality effort and suppliers tend to provide low-quality products because producing high-quality products costs more and the buyer cannot distinguish them even with inspection [31]. Several strategies can correct the problems associated with the moral hazard induced by asymmetric information in the supply chain. The most obvious strategy is to get more information about the supplier and the quality of the supplier’s product. This strategy will correct some of the information asymmetry, but acquiring accurate information is expensive and may be infeasible [32]. Another strategy is to expose the supplier to the liability risk of low product quality by adopting traceability systems [6]. A third strategy is to make revealing information valuable, encouraging the supplier to ‘signal’ its quality level in some fashion [11,32]. Quality signals include the adoption of process standards such as ISO 9001, HACCP compliance or guarantees, warranties and third-party certifications. The final strategy is to design contracts that appeal to high-quality suppliers but not to low-level-quality suppliers. The final strategy is less expensive than other strategies and can be used in all supply chains [11]. Thus, we herein focus on the contract-based strategy, developing an ontology-based approach to keep the principal–agent contract continually effective during its execution.

3. Principal–agent mathematical models and data-centric framework

3.1. Principal–agent mathematical models

Our ontology models are based on Starbird’s [15,31] principal–agent perspective. The conclusion is illustrated in equations (1) and (2). In equation (1), $p (θ^{H})$ denotes the probability of products passing inspection under the upstream supplier’s high-quality effort, $θ^{H}$ , and $p (θ^{L})$ denotes the probability under low-quality effort, $θ^{L}$ . Equation (2) explains that the probability level $p$ is determined by quality effort, $θ$ , inspection sampling size, $n$ , and inspection-accepted number, $d$

\frac{c (θ^{H}) - c (θ^{L})}{p (θ^{H}) - p (θ^{L})} \leq ω + s

(1)

p (θ) = \sum_{x = 0}^{d} (\begin{matrix} n \\ x \end{matrix}) {(1 - θ)}^{x} θ^{n - x}

(2)

where $ω$ in equation (1) corresponds to the price the downstream supplier (principal) pays to the upstream supplier (agent) for products, and $s$ is the penalty the upstream supplier pays to the downstream supplier if a product fails inspection. $c (θ^{H})$ and $c (θ^{L})$ are, respectively, the cost of producing products when the quality efforts are $θ^{H}$ and $θ^{L}$ . If the value of the right-hand side is greater than or equal to the left side of equation (1), the incentive-compatible conditions for the upstream supplier are satisfied. In that case, the upstream supplier is willing to exert a high-quality effort. Equation (2) demonstrates an indirect relationship between the sampling inspection policy and contract efficacy; the inspection policy can also be modified to produce a motivation effect. Consider this example to further explain equations (1) and (2). If $c (θ^{H})$ increases much, the inequality will not be met, so managers should increase $ω$ or $s$ , or the sampling size, $n$ , and $d$ to reestablish the relationship. The diagnostic error of the inspection tool in the inspection policy is not included in equation (2), but it should also be considered. We include this error in the updated equation (2) in section 5.2.

3.2. The data-centric framework

The data-centric framework was introduced by IBM and has been used to achieve substantial savings when in business transformations [33,34]. We used the framework in this study to design algorithms for identifying quality-risky suppliers, products and efficacy-losing contracts. The basic definitions and notations, and the model of the data-centric framework, are described in this section [35]. The basic definitions include artefact, artefact schema and service.

Definition 1

Artefact class: artefacts represent real or conceptual key business entities [36], defined as carrying attribute-record relations and internal-state relations that services can consult and update. An artefact class is a pair, $C = 〈 R, S 〉$ , where $R$ and $S$ are two relation symbols. An instance of $C$ is a pair $C = 〈 R, S 〉$ where (1) $R$ , called the attribute relation, is an interpretation of $R$ containing exactly one tuple over $D$ ( $D$ is the active domain) and (2) $S$ , called the state relation, is a finite interpretation of $S$ over $D$ .

Definition 2

Artefact schema: an artefact schema is a tuple, $A = 〈 C_{1}, \dots, C_{n}, DB 〉$ , where each $C_{i}$ is an artefact class, $DB$ is a database schema; and $C_{i}$ , $C_{j}$ and $DB$ have no relation symbol in common for $i \neq j$ . An instance of an artefact schema $= 〈 C_{1}, \dots, C_{n}, DB 〉$ , is a tuple $A = 〈 C_{1}, \dots, C_{n}, DB 〉$ , where $C_{i}$ is an instance of $C_{i}$ and DB is an instance of $DB$ over $D$ . By slight abuse, artefact schema $A$ can be identified as $A = DB \cup {R_{i}, S_{i} | 1 \leq i \leq n}$ . An instance of an artefact schema is a tuple of class instances, each corresponding to an artefact class plus a database instance.

Definition 3

Service class: a service, $σ$ , over an artefact schema $A$ is a tuple $σ = 〈 π, φ, Sr 〉$ ; where $π$ , called a precondition, is a sentence in $L_{A}$ . $L_{A}$ is the semantic formulas over $A$ . $φ$ , called a postcondition, is a formula in $L_{A}$ with free variables ${\bar{x}}_{R} | R$ , attribute relations of classes in $A$ . $Sr$ is a set of state rules containing for $S$ in $A$ both or none of the following rules

\begin{matrix} S (\bar{x}) \leftarrow \emptyset_{S}^{+} (\bar{x}) \\ \neg S (\bar{x}) \leftarrow {\emptyset_{S}}^{-} (\bar{x}) \end{matrix}

A service, $σ$ , transforms the instance $A$ to $A'$ , where $A'$ is another instance of $A$ . The transformation process can be formed as $A \binom{σ}{\to} A'$ .

An artefact system is guarded decidable if all formulas used in the state rules of its services are guarded; the service rules in this article are guarded according to the principles in Deutsch et al. [36].

4. The ontology-based approach

The ontology-based approach aims to monitor and assure the efficacy of the principal–agent contract. The approach consists of ontology-based models and data-centric algorithms.

4.1. The ontology-based models

Ontology is a popular approach to representing domain knowledge [37 –39]. Ontology models are developed in this study to represent concepts and relations between concepts involved in verifying whether suppliers incentive-compatible conditions are satisfied, or whether the suppliers are willing to exert high-quality effort during principal–agent contract execution. This will be so if the principal–agent contract is effective during execution. This research adopted the ontology framework developed by Jurisica et al. [40] consisting of four broad ontological categories that, respectively, deal with static, social, intentional and dynamic aspects of the world. For a wide range of real-world applications, the relevant knowledge can be represented based on the primitive concepts and relations from these four ontological categories. Static ontology describes the static aspects of the world (i.e. what things exist, their attributes and relationships). Social ontology describes the social relations in the world. Intentional ontology can model individual or organisational motivations, intentions, goals, beliefs, choices and so on. Dynamic ontology describes the changing aspects of the world in terms of states, state transitions and processes. These ontologies have been widely used for knowledge management. In this article, ontologies are defined as sets of relationship knowledge rules and data tuples of entities’ attributes.

4.1.1. Ontology metaclasses

To develop the ontology, we first propose four metaclasses: Organisation Class, Resource Class, Task Class and Goal Class. Every entity in the domain can be an instance of one of these metaclasses.

An organisation in the supply chain has strategic goals, observes and manipulates resources, and intentionally acts according to the principles of rationality within the organisational setting. Resources represent the data an organisation observes and the objects it can take action on including the inspection policy, contract and product as subclasses. The description of the resource status forms a proposition.

Goals represent an organisation’s strategic interests that refer to the actor’s desired state. Tasks represent the particular courses of action that can be executed to reach a goal.

We develop static, social, intentional, dynamic ontologies using the four metaclasses for supply-chain quality control. These four ontological categories can be specified as relationship constraints and attribute tuples as defined in Figure 1.

Figure 1.

Relations between metaclasses.

4.1.2. Static, social, intentional and dynamic ontologies

In this article, an ontology model is defined as sets of relationship knowledge rules and data tuples of objects, declaring the entities’ relationships and attributes in the supply-chain quality-control domain

\begin{matrix} O_{k} = {k | k is a knowledge rule declaring the relationships between entities} \\ O_{d} = {d | d is a data tuple declaring an entity' s attributes} \end{matrix}

Definition 4

Knowledge rules metadata: for each knowledge rule, $k \in O_{k}$ , its metadata is defined as a four tuple, $〈 kid, TY, P, H 〉$ , where $kid$ is the unique identifier for the knowledge rule. TY ∈ {static ontology, intentional ontology, social ontology, dynamic ontology}, $P$ is the premise of the knowledge rule and $H$ is the conclusion of the knowledge rule. Consider an example from Figure 4: $k 1 : (s 2, g 1) (s 2, g 2) (s 2, g 3)$ is a knowledge rule meaning that if supplier s2’s goal g1 exists, then its goal $g 2$ or g3 exists. Here, $k 1$ is the unique identifier of this knowledge rule. $k 1 . TY$ means the ontology type of the underlying knowledge rule k1 (i.e. intentional ontology). $k 1 . H = (s 2, g 2) \lor (s 2, g 3)$ , as the conclusion of the knowledge rule, is inferred by the premise of the rule $k 1 . P = (s 2, g 1)$ .

Definition 5

Object metadata: for each $d \in O_{d}$ , its metadata is defined as a tuple consisting of an entity’s attributes. Its instance is the data record for this object’s attributes. For example, a supplier is associated with attributes tuple $〈 d 1 : s_id, location 〉$ . Product is an entity associated with attributes tuple $d 2 : 〈 product_id, high_quality_\cos t, low_quality_\cos t, offered_price, inspection_results 〉$ :

Static ontology: static ontology represents the static aspect of quality control in the supply chain, defining entities’ attributes and relationships among entities. Entities in the static ontology include organisations, information, contract, and product. These entities’ attributes are modelled as data tuples and the relationships between them are modelled as knowledge rules. The resource status may affect organisations’ status. For example, news of increased product-material (r4) cost will increase the product’s (r5) manufacturing cost and further trigger the supplier’s (s2) stable-profit goal: $k 2 : r 4 \Rightarrow r 5$ and $k 3 : r 5 \Rightarrow s 2$ . The attributes metadata for the supplier is defined as $d 1 : 〈 s_id, location, product_id 〉$ , the attributes metadata for the product is defined as $d 2 : 〈 product_id, high_quality_\cos t, low_quality_\cos t, offered price, inspection_results 〉$ (Figure 2).

Social ontology: social ontology describes the social aspects of supply-chain quality control. In particular, it expresses the suppliers’ principal–agent interactions. For example, if supplier s2 who provide materials (r5) is an agent to supplier s3, and s3 is the principal, there is a contract (r3) between them. The knowledge rules can be defined as $k 4 : r 3 \subseteq s 1 \times s 2$ and $k 5 : r 5 \subseteq s 2 \times s 1$ (Figure 3).

Intentional ontology: intentional ontology models each organisation’s (e.g. supplier) motivation: what the supplier desires or intends to do. The supplier’s goals can be broken down into subgoals by AND and OR decompositions. For instance, when some events that will degrade profit arise, the goal of stable profit will be triggered. It can be decomposed into reducing production cost or increasing production offered price: $k 1 : (s 2, g 1) \Rightarrow (s 2, g 2) \lor (s 2, g 3)$ . A means-ends analysis can connect the organisation’s goals and activities in supply-chain quality control. For example, the goal of decreasing cost can be achieved by switching to low-quality production, which can be represented as a knowledge rule: $k 6 : (s 2, g 3) \Rightarrow (s 2, t 1)$ (Figure 4).

Dynamic ontology: dynamic ontology defines how resource sentences (resource sentence is a proposition describing a resource) trigger the goals and defines the state after carrying out tasks. In other words, the dynamic ontology represents the changing aspect of the supply-chain quality-control domain. For example, the goal of a stable profit is triggered by the sentence that the cost of product materials rises, which can be represented as knowledge rule $k 7 : sen (r 4) \Rightarrow (s 2, g 1)$ (Figure 5).

Figure 2.

Static ontology.

Figure 3.

Social ontology.

Figure 4.

Intentional ontology.

Figure 5.

Dynamic ontology.

4.1.3. Ontology development for supply-chain quality control

To ensure the scalability, generality and adequacy of the supply-chain quality-control ontologies, we propose an ontology-development method based on the four basic types of ontologies described in section 4.1.1. This method aims to guide users in developing their own supply-chain quality-control ontologies using those four basic types.

To assess whether the ontologies developed by a user appropriately cover the complete domain of interest (in supply-chain quality control), a common approach is to use competency questions – a natural language processing technique – to find relevant terms for ontology development. Based on the four ontology types of supply-chain quality control, we develop a set of general competency questions [41,42] to help supply-chain quality-control users develop ontologies for their specific needs:

Competency questions for key classes:

What kinds of organisations/suppliers are involved in quality control for this supply chain?

What kinds of data (records of information) and knowledge are interrelated with quality-effort behaviours?

What are the suppliers’ quality-effort tasks in the supply chain, and what is the goal for each task?

Competency questions for static ontology:

What relationships between resources and suppliers are helpful for tracking the supplier’s quality-effort motivation and tasks?

What attributes of resources, suppliers, and tasks are helpful for monitoring quality-effort motivation and tasks?

Competency questions for social ontology:

What kind of principal–agent contract exists for up- and downstream suppliers?

What products do the upstream suppliers provide to downstream suppliers?

What suppliers will be influenced if a goal is achieved?

Competency questions for intentional ontology:

What goal is the supplier committed to achieving?

What task must a supplier perform to achieve the goal?

Competency questions for dynamic ontology:

In which situation may the supplier act?

Is it possible for a supplier to perform a task in some situation? Does the supplier have the ability to perform the task?

What sequence of activities must be completed to achieve each goal?

4.2. Data-centric algorithms

4.2.1. The artefact and service classes

As noncontract parameter fluctuations are unobservable, we collect information of factors that determine noncontract parameters in the approach in this study. The information is seen as resources. Its proposition description is seen as resource sentences. The information is usually stored in text such as news, cooperation reports and marketplace databases. The information is collected and mapped to concepts and relations in the ontology-based models. With the mapped information as input, algorithms developed in the data-centric framework are used to identify quality-risky suppliers, risky products and efficacy-losing contracts. The artefact classes are the entities to be monitored: suppliers, products and contracts. The service classes define how these artefact classes change from one quality-related state to another. The artefact and service classes used in designing the algorithms are defined in this section.

4.2.2. Artefact classes

$Supplier = 〈 R_{s}, event_trigger, g, t 〉$ is the artefact carrying the information about a supplier:

$R_{s} (s_id, location, product_id)$ is the supplier’s attribute relation and $location$ is supplier’s location.

$event_trigger$ is a nullary state relation. If there is a relationship between the sentence (resource) and the supplier, the $event_trigger$ is filled. Otherwise, it is empty.

$g (s_id, goal)$ is a state relation. If sentences (resources) invoke a supplier’s goal in the dynamic ontology, then $g (s_id, goal)$ is filled.

$t (s_id, task)$ is a state relation. If relationships exist between a supplier’s goal and some quality-related production behaviour, then $t (s_id, task)$ is filled.

$\Pr oduct = 〈 R_{p}, idle, not_recall, quality_level, risk, inspected 〉$ :

$R_{p} (product_id, s_id, high_cost, low_cost, inspection_results)$ is the product’s attribute relation. $high_cost$ is the attribute recording the cost of producing a high-quality product, and $low_cost$ records the cost of producing a low-quality product.

idle is a nullary state relation. If the product is not monitored, $idle$ is filled.

$recall$ and $not_recall$ are nullary state relations. If the product passes inspection, $not_recall$ is filled; otherwise, $recall$ is filled.

$quality_level (product_id, θ)$ is a state relation. If a relationship exists between a supplier’s task and the task effect of production quality in the dynamic ontology, the state will be filled.

$risk (product_id)$ is a nullary state representing that the product is quality risky.

$inspected$ is a nullary state representing that whether the downstream supplier will inspect product quality.

$Contract = 〈 R_{c}, efficient, not_efficient 〉$ :

$R_{c} (contract_id, s_id, product_id, penalites, reward, accept_rate)$ is the attribute relation of the principal–agent contract.

If suppliers involved in the contract have a low-quality task and the product involved is quality risky, then the contract is filled by state $not_efficient$ ; otherwise, it is filled by the $efficient$ state; $not_efficient$ and $efficient$ are both nullary state relations.

4.2.3. Service classes

In supply-chain quality control, the service class reads sentences of resources and artefacts’ states and also changes artefacts’ states:

$〈 check_triggered_supplier = π^{ch_t_s}, ψ^{ch_t_s}, S r^{ch_t_s} 〉$ : Table 1 describes the services.

Table 1.

check_triggered_supplier.

Service description
$π^{ch_t_s} = sen (r)$
$ψ^{ch_t_s}$ is given by
$\begin{matrix} R_{s} (s_id, location, product_id) : = \exists r sen (r) \land \exists ki d_{1} k (ki d_{1}, ″ static ontology ″, r, product_id) \land \\ \exists ki d_{2}, s_i d_{1} k (ki d_{2}, ″ social ontology ″, product_id, s_id \times s_i d_{1}) \land \\ \exists di d_{1} d (di d_{1}, supplier, s_id, location) \end{matrix}$
$S r^{ch_t_s}$ is given by
$event_trigger \leftarrow true$

It applies over the supplier artefact when $sen (r)$ is true and fills attributes relation $R_{s}$ for suppliers who may be affected by sentences of resource $r$ :

$check_supplier_goal = 〈 π^{ch_s_g}, ψ^{ch_s_g}, S r^{ch_s_g} 〉$ : Table 2 describes the services.

Table 2.

check_supplier_goal.

Service description
$π^{ch_s_g} = event_trigger$
$S r^{ch_s_g}$ is given by
$g (s_i d_{,} goal) \leftarrow \exists locatio n_{,} product_id R_{s} (s_i d_{,} locatio n_{,} product_id) \land$
$\exists ki d_{3,} product_id sen (product_id) \land k (ki d_{3,} dynamic ontologysen (product_id), goal,)$
$t (s_i d_{,} task) \leftarrow \exists locatio n_{,} product_id R_{s} (s_i d_{,} locatio n_{,} product_id) \land \exists goal g (s_i d_{,} goal) \land$
$\exists k i d_{4} k (k i d_{4}, " i n e n t i o n a l o n t o l o g y, " (s_i d_{,} g o a l), (s_i d_{,} t a s k))$

$product_initiate = π^{pro_ini}, ψ^{pro_ini}, S r^{pro_ini}$ : Table 3 describes the service.

Table 3.

product_initiate.

Service description
$π^{pro_ini} = \exists s_id, location, product_id R_{s} (s_id, location, product_id) \land \exists goal, task t (s_id, goal, task) \land idle$ $ψ^{pro_ini}$ is given
$\begin{array}{l} R_{p} (p r o d u c t_i d, s_i d, h i g h_c o s t, l o w_c o s t, ω) ∶ = R_{p} (ω, ω, ω, ω, ω) \land \\ \exists location R_{s} (s_i d, l o c a t i o n, p r o d u c t_i d) \land \\ by \exists product_i d, d i d_{2} d (d i d_{2}, p r o d u c t_i d, h i g h_c o s t, l o w_c o s t) \land \\ high_cost > ω \end{array}$
$S r^{p r o_i n i}$ is given by
$\begin{array}{l} quality_level (p r o d u c_i d_{,} θ^{'}) \leftarrow \exists s_i d t (s_i d, " s t a b l e p r o f i t ") \geq θ^{'} \geq ω \land \\ \exists d i d_{3}, c o n t r a c t_i d, p e n a l i t e s, r e w a r d s, a c c e p t_r a t e d \\ (contract_i d, p r o d u c t_i d, s_i d, p e n a l i t e s, r e w a r d, a c c e p t_r a t e) \land \\ \exists s_i d, p r o d u c t_i d, h i g h_c o s t, l o w_c o s t R_{p} (s_i d, p r o d u c t_i d, h i g h_c o s t, l o w_c o s t) \land \\ principal_agent_modle (h i g h_c o s t, l o w_c o s t, p e n a l t i e s, r e w a r d s) < 0 \\ i n s p e c t e d \leftarrow true \end{array}$

Here, the manufacturer’s cost is recorded under high-quality production and low-quality production, respectively.

$inspect = π^{ins}, ψ^{ins}, S r^{ins}$ : Table 4 describes the service.

Table 4.

Inspect.

Service description
$π^{ins} = inspected$
$ψ^{ins}$ is given by
$\begin{matrix} R_{p} (product_id, s_id, high_cost, low_cost, θ ″) ~ = \exists product_id, s_id, high_cost, \\ low_cost R_{p} (product_id, s_id, high_cost, low_cost, ω) \land \\ θ ″ = accept_model (n, d, tool, quality_level (θ')) \land \\ \exists di d_{4}, n, d, tool d (inspection_id, n, d, tool) \end{matrix}$
$S r^{ins}$ is given by
$risk (product_id) \leftarrow \exists θ', θ' compare (inspect_results (θ ″), quality_level (product_id, θ')) > 0$

$check_contract = π^{ch_c}, ψ^{ch_c}, S r^{ch_c}$ Table 5 describes the service.

Table 5.

check_contract.

Service description
${\begin{matrix} π \end{matrix}}^{ch_c} = efficient$
${\begin{matrix} ψ \end{matrix}}^{ch_c} = is given by$
$\begin{matrix} R_{c} (contract_id, s_id, product_id, penalites, rewards, accept_rate) \\ : = \exists di d_{5}, d (di d_{5}, ″ contract_id ″, s_id, product_id, penalites, rewards, accept_rate) \land \\ \exists ki d_{5}, s_i d_{2} k (ki d_{5}, ″ socialontology, ″ contract_id, s_id, s_i d_{2}) \end{matrix}$
$\begin{matrix} not_efficient \leftarrow \exists product_idrisk (product_id) \land \exists s_id R_{s} (s_id, product_id) \land \exists c R_{c} () \\ not_efficient \leftarrow \exists product_idrisk (product_id) \land \exists s_id R_{s} (s_id, product_id) \land \\ \exists contract_id, penalites, rewards, accept_rate R_{c} (contract_id, \\ s_id, product_id, penalites, rewards, accept_rate) \end{matrix}$

4.2.4. Risky suppliers/product materials and losing efficacy contracts identification algorithms

Using the artefacts and services classes defined in section 4.2.1, we design the algorithms for identifying quality risky supplier, risky product and efficacy-losing contract.

Definition 6

Quality risky product: when state $risk (production_id)$ of a product is set as true, the product is of low quality.

Definition 7

Quality-risky suppliers of a product: when a product is quality risky, then its supplier is quality risky. $Risky supplier = {s_id | R_{p} (s_id, \dots, production_id, \dots, \dots) \land risk (production_id)}$ .

Based on the definitions of risky suppliers and products, Table 6 provides an algorithm for identifying them.

Table 6.

Risky product materials/suppliers identification algorithms.

Algorithm steps
Step 1. Set the event as the current event. Step 2. Apply services including check_triggered_supplier, check_supplier_goal, product_initiate, inspect sequentially. Step 3. For every instance of artefact class Production, if state risk (production_id) is set true, then add the production_id into a quality-risky production set ${product_id \| risk state of the product is set true}$ . The s_id is added to a risky-supplier set ${s_id \| R_{p} (s_id, \dots, production_id, \dots,) \land risk (production_id)}$ .

Algorithm steps

Step 1. Set the event as the current event.
Step 2. Apply services including check_triggered_supplier, check_supplier_goal, product_initiate, inspect sequentially.
Step 3. For every instance of artefact class Production, if state risk (production_id) is set true, then add the production_id into a quality-risky production set

{product_id | risk state of the product is set true}

.
The s_id is added to a risky-supplier set

{s_id | R_{p} (s_id, \dots, production_id, \dots,) \land risk (production_id)}

Definition 8

Efficacy-losing principal–agent contract: if the supplier and product involved in a principal–agent contract are, respectively, included in the risky-supplier and risky-product sets, the contract is losing efficacy

contract (losing efficacy) = {contract (s_id, product_id) | s_id \in risky supplier set \land product_id \in risky product set}

Based on the definition of an efficacy-losing principal–agent contract, Table 7 provides an algorithm for identifying them.

Table 7.

Efficacy-losing contract identification algorithm.

Algorithm steps
Step 1. Find the suppliers and products in the contract. Step 2. Check whether the suppliers and products involved in the contract are, respectively, in the risky-suppliers or risky-product set. Step 3. If the suppliers and product are, respectively, in the risky-supplier and risky-product sets, then set the contract as losing efficacy and encourage the downstream supplier to modify contract parameters.

Algorithm steps

Step 1. Find the suppliers and products in the contract.
Step 2. Check whether the suppliers and products involved in the contract are, respectively, in the risky-suppliers or risky-product set.
Step 3. If the suppliers and product are, respectively, in the risky-supplier and risky-product sets, then set the contract as losing efficacy and encourage the downstream supplier to modify contract parameters.

5. Evaluation

In this section, we evaluate the ontology-based approach proposed in this study through case study, simulation and comparison against other approaches related to supply-chain quality control.

5.1. Case study

To prove the appropriateness of the ontology-based approach proposed in this research, we report a case study in this section.

5.1.1. Milk supply-chain quality-control description

We conduct a case study in the milk supply-chain domain. In this supply chain, one of the biggest, most famous milk suppliers (named milk company A here for simplicity) buys raw milk from producers distributed in different places (raw milk producers are dairy farmers). Some dairy farmers are located in an area called Sichuan, where rain can continue for days. The climate is very humid at that time, so hay for feeding milk cows is easily destroyed during that time, easily leading to excess Aspergillus enzyme in the raw milk. The cost of feeding milk cows high-quality hay (high-quality effort cost) is higher than in places where the climate is dry.

Unfortunately, it is difficult for milk company A to determine whether the raw milk has excess Aspergillus enzyme. The inequality in equation (1) will not be established for the dairy farmers; thus, the suppliers located in humid places choose to feed cows the destroyed hay. Then, risky product and suppliers will appear in the supply chain.

5.1.2. Ontology development

As introduced in section 4.1.3, the user may start the ontology development using the competency questions. After the user answers the competency questions according to Figure 6, the milk manufacturers and dairy farmers are identified as organisations, with goals of stable profit and related tasks. Milk company A, as a downstream supplier, buys raw milk from dairy farmers (upstream suppliers). Thus, the raw milk product, the contract between the suppliers, and some information about the weather form the ontology resource classes. Static, social, intentional and dynamic ontology models are developed, respectively (Figures 7 –10).

Figure 6.

Ontology-development competency questions.

Figure 7.

Static ontology of milk supply-chain quality control.

Figure 8.

Social ontology for milk supply-chain quality control.

Figure 9.

Intentional ontology for milk supply-chain quality control.

Figure 10.

Dynamic ontology for milk supply-chain quality control.

Static ontology of the milk supply chain. In the static ontology, the dairy farmers supply milk manufacturers with raw milk, which is the most important material of the milk product. The quality level of the raw milk largely affects the quality of the milk product.

$k_{c 1} : Sichuan dairy farmers \subseteq raw milk suppliers$ . In $k_{c 1}$ , Sichuan dairy farmers form a subclass of the dairy farmers class. If the cost of high-quality raw milk increases, which is an attribute of r4, supplier s3’s goal is triggered and their quality effort behaviours may be affected. $k_{c 2} : r 4 \Rightarrow s 3$ , attributes of milk may affect milk manufacturers’ (s2) goal $k_{c 3} : r 1 \Rightarrow s 2$ .

Data tuple s3, Sichuan dairy farmers, is described as $〈 d_{c 1} : s_id, Sichuan, product_id 〉$ and data tuple r4: raw milk is described as $〈 d_{c 2} : production_id, c_{H}, c_{L}, pric e_{H}, inspection_results 〉$ . Data tuple contract is $〈 d_{c 3} : milk_manu, dairy_farmer, penalties, rewards, acc_rate, pric e_{H} 〉$ , $milk_manu$ is the principal supplier in the contract, $dairy_farmer$ is the agent supplier in the contract, $penalties$ and $rewards$ are the motivation provisions, $acc_rate, pric e_{H}$ is the acceptance rate for raw milk after inspection and $pric e_{H}$ is the offered price for accepted raw milk.

Social ontology for milk supply-chain quality control. The milk manufacturer, as a principal, designs a contract. The dairy farmers, as agents, determine whether to accept the contract or not. Once accepted, a contract relation exists between them. The milk manufacturer receives raw milk produced by a dairy farmer. Thus, a raw milk supply relation exists between them: $k_{c 4} : contract \subseteq s 2 \times s 1$ and $k_{c 5} : raw milk \subseteq s 1 \times s 2$ .

Intentional ontology for milk supply-chain quality control. Once the cost of producing raw milk without excess Aspergillus enzyme increases, the diary farmer’s intentional states may be affected: The goal of ‘stable profit’ and its subgoal ‘decrease cost’ may be triggered and the means-end for ‘feeding_cured_hay’ may take effect $k_{c 6} : g 1 \Rightarrow g 2 \lor g 3$ and $k_{c 7} : g 3 \Rightarrow t 1$ .

Dynamic ontology for milk supply-chain quality control. The dynamic ontology defines the resource sentences (propositions) that will trigger the goals and that specify the results of tasks. It represents the changing aspect of the supply-chain quality-control domain. $k_{c 8} : φ_{1} (r 4) \Rightarrow g 3$ , $k_{c 9} : g 3 \Rightarrow t 1$ , $k_{c 10} : t 1 \Rightarrow φ_{2} (r 4)$ , $φ_{1} (r 4)$ and $φ_{2} (r 4)$ are, respectively, the sentences about the cost and quality level of raw milk, and how the cost and quality are attributes of raw milk.

5.1.3. Quality-risky suppliers and products and efficacy-losing contract identification

The algorithms designed for identifying risky products and suppliers, and efficacy-losing contracts are in section 4.2. Supplier, product and contract are defined as data-centric artefacts; the approach proposed in this research aims to monitor the quality-related states of the three artefacts. The artefact states can be consulted and updated using the ontology models as a database.

For the identification of quality-risky suppliers and products and efficacy-losing contract, the algorithms in section 4.2 apply services ‘check_triggered_supplier’ on the static- and social-ontology bases, ‘check_supplier_goal’ on the intentional ontology base. Then ‘product_initiate’ and ‘inspect’ services are applied to the product artefact. The final states of suppliers, products and contract are listed in Table 8. Because raw milk’s risk state is filled, the product and its supplier are quality risky. The contracts involving risky suppliers and products loses efficacy.

Table 8.

The postcondition and states of suppliers, product and contract.

The final states description
Raw milk supplier:
$\begin{matrix} R_{s} (s 2, ″ Sichuan ″, r 4) \\ g (s 2, ″ decreasecost ″) \\ t (s 2, ″ feeding_cured_hay ″) \end{matrix}$
Raw milk:
$\begin{matrix} R_{p} (r 4, s 2, c_{H}, c_{L}, ω) \\ quality_level \leftarrow θ' \\ inspect_results \leftarrow θ ″ \\ risk (r 4) (if θ' > θ ″) \end{matrix}$
Contract:
$R_{c} (s 2, r 4, rewards, penalties, p (θ))$
inefficient

5.2. Comparison against traditional principal–agent contract-based approach using simulation data

The efficacy of a principal–agent contract depends not only on contract parameters such as offer price, penalties, rewards, sampling size and acceptance number but also on factors not included in contract such as cost of producing a high-quality product, capacity and the cost of the inspection policy. The factors not included in the contract usually fluctuate even during the contract period, and their fluctuations are usually unobservable, which may result in contract-efficacy failure. The ontology-based approach is developed to capture these fluctuations and use them to identify quality-risky suppliers and products, which in turn determine whether the contract is effective. By identifying efficacy-losing contracts, the downstream supplier issuing the contract can adjust the contract’s motivation parameters or inspection policy to ensure its efficacy. The ontology-based approach is a tool enabling the principal–agent-contract approach to motivate upstream suppliers to exert a high-quality effort.

The combination of our approach with principal–agent contracts produces a dynamic principal–agent-contract approach while traditional principal–agent contract without our approach is a static principal–agent-contract approach. In this section, we compare dynamic principal–agent contracts against static principal–agent contracts in terms of the downstream supplier’s total utility by assuming that it prefers the upstream supplier exerts a high-quality effort. In this evaluation, we use simulation data in which we assume that the cost of producing a high-quality product fluctuates stochastically. For the dynamic approach, the downstream supplier notices the fluctuations of the high-quality effort and adjusts contract parameters. There are two alternatives, one adjusts the offer price for the product passing inspection and the other adjusts the inspection-policy parameters including sampling size, acceptance number and diagnostic error of the inspection tool adopted in the inspection policy. In the static approach, the downstream supplier does not notice and makes no change. We compare the approaches in terms of the downstream supplier’s utility.

To better explicate the simulation process and comparison results, we explain the mathematical models for computing the probability rate of much product passing inspection at a certain level of quality, computing the total utility of the downstream supplier who chooses to motivate upstream suppliers to exert a high-quality effort and computing the incentive-compatible conditions to motivate upstream suppliers to exert a high-quality effort. These models are derived from Starbird’s [15] work. Some notations herein are common to those in section 3.1, where they have the same meaning.

5.2.1. Related mathematical model descriptions

The model for computing the acceptance rate under quality level $θ$ .

If $m$ is a standard established by the government or a customer, then high effort leads to a higher probability of satisfying that standard, but it does not guarantee that the standard is satisfied. We define $θ^{H}$ as the probability that a unit conforms to the standard under the high-quality effort $θ^{H} = F (m, e^{H})$ and $θ^{L} = F (m, e^{L})$ as the probability that a unit conforms to the standard under low effort. $e^{H}$ and $e^{L}$ are the high-quality effort and low-quality effort, respectively. The downstream supplier cannot directly observe the quality level of the product delivered by the upstream supplier. Instead, the downstream supplier uses a sampling plan to inspect the lot and to decide whether to accept it. The sampling plan involves drawing a random sample of size $n$ from a lot of size $N (N >> n)$ . If the number of low-quality units in the sample is less than or equal to the acceptance number, $d$ , then the entire lot is accepted and paid for. Otherwise, the entire lot is rejected. The probability that the lot is accepted is $p (θ)$ , so $θ$ is the probability that a unit in the lot satisfies standard $m$ ( $0 < θ \leq 1$ , $0 < p (θ) \leq 1$ , $p' > 0$ ). If we assume that every unit in the lot has the same probability of satisfying the standard, the acceptance probability rate is computed as

p (θ) = \sum_{x = 0}^{d} (\begin{matrix} n \\ x \end{matrix}) {(1 - θ)}^{x} θ^{n - x}

(3)

where $x$ is the number of units in the sample that do not pass inspection, $n$ is the sample size and $d$ is the acceptance number. The probability that a lot is accepted depends on the quality effort, and the parameters of the sampling plan – the sample size $n$ and the acceptance number $d$ . In addition, the probability of a lot being accepted is also affected by the diagnostic error of the inspection tool. Thus, we rewrite equation (3) into equations (4) and (5)

p (θ^{H}, ε) = \sum_{x = 0}^{d} (\begin{matrix} n \\ x \end{matrix}) {(1 - θ^{H} (1 - ε))}^{x} {(θ^{H} (1 - ε))}^{n - x}

(4)

p (θ^{L}, ε) = \sum_{x = 0}^{d} (\begin{matrix} n \\ x \end{matrix}) {(1 - θ^{L} (1 + ε))}^{x} {(θ^{L} (1 + ε))}^{n - x}

(5)

$ε$ , $0 < ε \leq (θ^{H} - θ^{L}) / 2$ , reflects the extent of the inspection tool’s diagnostic error. It decreases the probability of a high-quality product passing inspection and increases the probability of a low-quality product passing inspection. When $θ^{H} (1 - ε) = θ^{L} (1 + ε)$ , or $ε = (θ^{H} - θ^{L}) / 2$ , the inspection tool is unable to distinguish high quality and low quality.

The model for computing the downstream supplier’s total utility.

Downstream suppliers are assumed to prefer high-quality products. They cannot directly observe upstream suppliers’ efforts, but they can choose contract parameters to motivate upstream suppliers to exert high-quality effort. Derived from standard work, the total utility can be computed as

Utilit y_{downstream} = p (θ^{H}, ε) V [π N - ω N - (1 - θ^{H}) \times γ N - c^{t} (ε) n] + (1 - p (θ^{H}, ε)) V [- c^{t} (ε) n]

(6)

$π$ is the product value of each unit in the lot, $N$ is the lot size, $ω is the offer$ price the downstream supplier providing to the upstream supplier and $γ$ is the external penalty when a unit passes inspection but it does not satisfy standard $m$ . $n$ has the same meaning in equation (5). $c^{t} (ε)$ is the cost of inspecting each unit with the inspection tool of diagnostic error $ε$ . $V$ is the utility function of the downstream supplier. To simplify, we set $V (y) = y$ , so equation (6) is rewritten as

Utilit y_{downstream} = p (θ^{H}, ε) \times (π N - ω N - (1 - θ^{H}) \times γ N - c^{t} (ε) n) + (1 - p (θ^{H}, ε)) \times (- c^{t} (ε) n)

(7)

The incentive-compatibility condition models for upstream suppliers.

Contract parameters, including offer price, penalty and sampling plan, and other noncontract parameters, such as cost of producing products of varying quality levels, affect the upstream suppliers’ quality-level choices. An upstream supplier will exert high-quality effort only if incentive-compatible conditions in equations (8) and (9) are satisfied

p (θ^{H}, ε) U (ω N) + (1 - p (θ^{H}, ε)) U (- sN) - Nc (θ^{H}) \geq \underline{U}

(8)

$U$ is the preserved utility. $s$ is an internal penalty when the products do not pass inspection. $c (θ^{H})$ is the cost of producing high-quality effort. $c (θ^{L})$ is the cost of producing low-quality product

p (θ^{H}, ε) U (ω N) + (1 - p (θ^{H}, ε)) U (- sN) - Nc (θ^{H}) \geq p (θ^{L}, ε) U (ω N) + (1 - p (θ^{L}, ε)) U (- sN) - Nc (θ^{L})

(9)

To simplify, we set $U (y) = y$ , therefore equations (8) and (9) are rewritten as

p (θ^{H}, ε) \times (ω N) + (1 - p (θ^{H}, ε)) \times (- sN) - Nc (θ^{H}) \geq \underline{U}

(10)

p (θ^{H}, ε) \times (ω N) + (1 - p (θ^{H}, ε)) \times (- sN) - Nc (θ^{H}) \geq p (θ^{L}, ε) \times (ω N) + (1 - p (θ^{L}, ε)) \times (- sN) - Nc (θ^{L})

(11)

5.2.2. Simulation

We assume that the downstream supplier prefers high effort and it prefers to motivate the upstream supplier to exert a high effort. In addition, we assume that the cost of producing high-quality products for the upstream supplier, $c (θ^{H})$ , fluctuates stochastically, and $c (θ^{L})$ is fixed, the value of $c (θ^{H}) - c (θ^{L})$ is consistent with the fluctuation of $c (θ^{H})$ and is the value of the abscissa in Figures 11 and 12. The principal–agent contract is initially efficient. When $c (θ^{H})$ fluctuates, the contract may fail. We compare the static- and dynamic-contract approaches in terms of the total utility of the downstream supplier. In the static-contract approach, the downstream supplier cannot notice the fluctuation information of the cost and will not adjust contract parameters. In the dynamic-contract approach, the downstream supplier does notices and has two alternatives, changing the offer price, $ω$ , and changing the inspection policy, including the sampling size $n$ , the acceptance number $d$ and the inspection tool’s diagnostic error $ε$ . Offer price and inspection-policy parameters are associated with subscript characters st and dy for the two contract approaches, respectively. The total utility of the downstream supplier obtained from the static and dynamic contracts are illustrated as the values of the ordinate in Figures 11 and 12. Figure 11 presents comparison results with changing offer price, $ω$ , while Figure 12 presents comparison results with changing inspection-policy parameters $n$ , $d$ and $ε$ . Other parameters are fixed, and their value is presented as figure notes.

Figure 11.

The comparison results with adjusting offer price. $π = 115$ , $N = 300$ , $γ = 154$ , $ε = 0.01$ , $c^{t} (ε) = 0.5$ , $s = 20$ , $θ^{H} = 0.91$ , $θ^{L} = 0.8078$ , $n = 13$ and $d = 3$ .

Figure 12.

The comparison results with adjusting inspection policy. $π = 115$ , $N = 300$ , $γ = 154$ , $ω = 40$ , $s = 20$ , $θ^{H} = 0.91$ and $θ^{L} = 0.8078$ .

The contract is efficient when it is at the initial state (referring to the time point when the contract begins to execute), and in the initial state, the difference $c (θ^{H}) - c (θ^{L})$ is $10. In Figures 11 and 12, when the difference $c (θ^{H}) - c (θ^{L})$ is greater than $10, the total utility of the downstream supplier from using the dynamic principal–agent contract (corresponding to blue line) is greater than using the static principal–agent contract (corresponding to red line). When the difference $c (θ^{H}) - c (θ^{L})$ is less than $10, the downstream supplier has the same utility in the two approaches. The comparison results from the two alternatives prove that our approach is more efficient than the traditional static principal–agent contract-based approach.

5.3. Comparison against other approaches in supply-chain quality control

Although many studies have addressed the issues of supply-chain quality control, our study contributes a new tool in this field. We compare our research against existing approaches. After a review of the literature in the field of supply-chain quality control, we extract eight dimensions for comparing our approach with others. These dimensions include type, product quality, product safety and security, expense, complex supply chain, moral hazard, asymmetric information and real time. Type refers to the tactics, principles or tools each approach belongs to. Product quality refers to whether these approaches have a direct target of improving product quality. Product safety and security refers to whether each approach can reduce the probability of adulteration and that use of a product will result in illness, injury, death, and negative consequences to people, property or equipment. Expense refers to whether each approach is costly. Complex supply chain refers to whether each approach is suitable in a complex supply-chain network. Moral hazard refers to whether the approach can handle moral hazard induced by asymmetric information. Asymmetric information refers to whether each approach can reduces asymmetric information. Real time refers to whether each approach can monitor the supplier’s quality effort in real time. The approaches to be compared are tactics involving promoting supply-chain communication and partnership activities [17,19], process integration and management of supply-chain linkages [20], management and leadership [43] and regulations and standards such as ISO 9001, QS-9000, total quality management and HACCP [16,21 –23]. Tools include traceability systems [6], principal–agent contracts [15] and our approach. Supply-chain communication and partnership activities involve organisations working closely together and nurturing relationships with other members of the supply channel to share goals and coordinate activities. Process integration and management of supply-chain linkages refers to a smooth and synchronised linkage between dissimilar processes and operations. Vertical integration of firms is aligned with this tactic. Management and leadership encourage supply-chain-member participation and cultivate quality measurement.

The tactics used in supply-chain quality control are meta-abstractive. Managers must transform them into specific operation-management activities using their management experience. This is a challenge for managers as they have different experiences. In addition, these types of approaches, especially process integration, are costly [34] and unsuitable in complex supply-chain networks. Regulations and standards have two types of roles, as guidelines for whole supply-chain operations and that downstream members in the supply chain prefer to choose upstream suppliers who operate in alignment with various regulations and standards [16]. Regulations and standards also have the meta-abstractive problem, as suppliers may pretend their product meets regulations and standards. As suppliers are often located in different places or countries, the regulations and standards may not be consistent across the supply chain, thus they are also unsuitable in complex supply chains. In addition, tactics as well as regulations and standards usually cannot handle moral hazard and asymmetric information.

Traceability systems and principal–agent contracts have been proposed to overcome moral-hazard problems by motivating suppliers to exert high-quality effort [6,15]. However, traceability systems are expensive. Principal–agent contracts are inexpensive [10], but contract parameters often lose motivational efficacy [15]. Our approach provides principal–agent contracts with an ontology-based approach for monitoring whether the contract has fail during execution, revealing information of the supplier’s quality effort. Then, the manager can be notified of the change of the contract parameters to ensure its efficacy in real time. The comparison results are listed in Table 9. From the table, we can see that the ontology-based approach is a novel tool complementary to other extant approaches in supply-chain quality control. Compared with other approaches, it can reduce moral hazard and asymmetric information. In addition, the tool is inexpensive and suitable in complex supply-chain networks.

Table 9.

The comparison results against other approaches.

Relatedapproaches	Supply-chaincommunicationand partnershipactivities	Processintegrationand supply-chainlinkagesmanagement	Managementand leadership	Regulations andstandards	Traceabilitysystems	Traditionalprincipal–agentcontract	Ourapproach
Type	Tactics	Tactics	Tactics	Principles	Tool	Tool	Tool
Product quality	✓	✓	✓	✓	✓	✓	✓
Product safetyand security	✓	✓	✓	✓	✓	✓	✓
Expense				✓		✓	✓
Complex supplychain						✓	✓
Moral hazard				✓	✓	✓	✓
Asymmetricinformation							✓
Real time							✓

6. Conclusion

The ontology-based approach proposed in this article is designed to make principal–agent contracts continually effective during contract execution. The approach consists of ontology-based models and data-centric algorithms. The ontology models are developed to represent concepts and relations between concepts involved in computing whether the contract effectively motivates suppliers to exert high-quality effort. The data-centric algorithms are designed to identify contracts losing efficacy using knowledge from the ontology-based models and timely multichannel data such as news and marketplace reports, which can be mapped to concepts or relations in the ontology-based models. This article’s contributions can be summarised as follows: First, it is one of the first attempts to monitor contracts losing efficacy during contract execution and notify contract issuers in time to change contract parameters to ensure the contract remains effective. Second, the approach can mitigate both asymmetric-information and moral-hazard problems. Third, data implicitly involving information related to suppliers’ quality effort can be collected and mapped to concepts and relations in the ontology-based models. This can help supply chains use timely data to control quality, which is uncovered by extant approaches. Finally, evaluations of case study, simulation and comparison with related approaches are conducted in this article. The evaluations of case study and simulation demonstrate that our approach is appropriate, can improve the efficacy of traditional principal–agent contracts and offer downstream suppliers higher utility from upstream suppliers’ high-quality effort. The evaluation against related extant approaches shows that our approach is a novel and inexpensive tool, and can be directly used in supply-chain quality control. In the future, we will develop related information systems and collect empirical data to verify our approach.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

This research is supported by the National Natural Science Foundation of China (Grant No. 71461023).

References

Albino

Carbonara

Giannoccaro

Supply chain cooperation in industrial districts: a simulation analysis. Eur J Oper Res 2007; 177: 261–280.

Balachandran

Radhakrishnan

Quality implications of warranties in a supply chain. Manage Sci 2005; 51: 1266–1277.

Pei

Tandon

Alldrick

et al . The China melamine milk scandal and its implications for food safety regulation. Food Policy 2011; 36: 412–420.

Stanciu

Stanciuc

Dumitrascu

et al . The effects of horse meat scandal on Romanian meat market, SEA. Pract Appl Sci 2013; 1: 174–181.

Zhang

. An evaluation of crisis response performance of Husi Food Co. during its 2014 food safety scandal – using the situational crisis communication theory. 2015, https://ruor.uottawa.ca/handle/10393/34903

Pouliot

Sumner

DA.

Traceability, liability, and incentives for food safety and quality. Am J Agr Econ 2008; 90: 15–27.

Aung

Chang

YS.

Traceability in a food supply chain: safety and quality perspectives. Food Control 2014; 39: 172–184.

Schwägele

Traceability from a European perspective. Meat Sci 2005; 71: 164–173.

Marucheck

Greisb

Menac

et al . Product safety and security in the global supply chain: issues, challenges and research opportunities. J Oper Manag 2011; 29: 707–720.

10.

Michael

McCathie

The pros and cons of RFID in supply chain management. Int Conf Mob Bus 2005; 721: 623–629.

11.

Starbird

SA.

Supply chain contracts and food safety. Choices 2005; 20: 123–128.

12.

Harris

Raviv

Optimal incentive contracts with imperfect information. J Econ Theory 1979; 20: 231–259.

13.

Sannikov

Moral hazard and long-run incentives. Working paper no. 3430, 22April2014. Standford, CA: Standford University.

14.

Prat

Jovanovic

Dynamic contracts when the agent’s quality is unknown. Theor Econ 2014; 9: 865–914.

15.

Starbird

SA.

Moral hazard, inspection policy, and food safety. Am J Agr Econ 2005; 87: 15–27.

16.

Robinson

Malhotra

MK.

Defining the concept of supply chain quality management and its relevance to academic and industrial practice. Int J Prod Econ 2005; 96: 315–337.

17.

Choi

Hartley

JL.

An exploration of supplier selection practices across the supply chain. J Oper Manag 1996; 14: 333–343.

18.

Forza

Filippini

TQM impact on quality conformance and customer satisfaction: a causal model. Int J Prod Econ 1998; 55: 1–20.

19.

Tan

Kannan

Handfield

et al . Supply chain management: an empirical study of its impact on performance. Int J Oper Prod Man 1999; 19: 1034–1052.

20.

Segars

Harkness

Kettinger

WJ.

Process management and supply-chain integration at the Bose Corporation. Interfaces 2001; 31: 102–114.

21.

Johnson

. Empirical study of second-tier automotive suppliers achieving QS-9000 Reference No. 718. Int J Oper Prod Man 2002; 22: 909–928.

22.

Terziovski

Power

Sohal

AS.

The longitudinal effects of the ISO 9000 certification process on business performance. Eur J Oper Res 2003; 146: 580–595.

23.

Kannan

Tan

KC.

Just in time, total quality management, and supply chain management: understanding their linkages and impact on business performance. Omega 2005; 33: 153–162.

24.

Uluskan

Joines

Godfrey

AB.

Comprehensive insight into supplier quality and the impact of quality strategies of suppliers on outsourcing decisions. Supply Chain Manag 2016; 21: 92–102.

25.

Tang

Pyke

How to mitigate product safety risks proactively? Process, challenges and opportunities. Int J Logist: Res App 2010; 13: 243–256.

26.

Kwon

I-WG

Kim

S-H

Martin

DG.

Healthcare supply chain management; strategic areas for quality and financial improvement. Technol Forecast Soc 2016; 113: 422–428.

27.

Lee

Whang

Higher supply chain security with lower cost: lessons from total quality management. Int J Prod Econ 2005; 96: 289–300.

28.

Luning

Devlierhere

Verhe

Safety in the agri-food chain. Wageningen: Wageningen Academic Publishers, 2006.

29.

Tang

Making products safe: process and challenges. Int Commer Rev 2008; 8: 48–55.

30.

Doherty

Smetters

Moral hazard in reinsurance markets. J Risk Insur 2005; 72: 375–391.

31.

Starbird

SA.

Penalties, rewards, and inspection: provisions for quality in supply chain contracts. J Oper Res Soc 2001; 52: 109–115.

32.

Hölmstorm

Moral hazard and observability. Bell J Econ 1979; 10: 74–91.

33.

Cohn

Hull

Business artifacts: a data-centric approach to modeling business operations and processes. IEEE Data Eng Bull 2009; 32: 3–9.

34.

Bhattacharya

Hull

A data-centric design methodology for business processes. Handb Res Bus Process Model 2009; 503–531.

35.

Baghdadi

A business model for deploying Web services: a data-centric approach based on factual dependencies. Inf Syst E-bus Manag 2005; 3: 151–173.

36.

Deutsch

Hull

Patrizi

et al . Automatic verification of data-centric business processes. In: Proceedings of the 12th international conference on database theory, Saint Petersburg, 23–25 March 2009, pp. 252–267. New York: ACM.

37.

Yan

Zhao

et al . Ontology-based scenario modeling and analysis for bank stress testing. Decis Support Syst 2014; 63: 81–94.

38.

Wang

Yan

et al . Ontologies for crisis contagion management in financial institutions. J Inf Sci 2009; 35: 548–563.

39.

Kittler

Towards an ontology of media. J Inf Sci 2017; 26: 23–31.

40.

Jurisica

Mylopoulos

Ontologies for knowledge management: an information systems perspective. Knowl Inf Syst 2004; 6: 380–401.

41.

Velardi

Navigli

Cuchiarelli

et al . Evaluation of OntoLearn, a methodology for automatic learning of domain ontologies. In: Buitelaar

Cimiano

Magnini

(eds) Ontology learning from text: methods, evaluation and applications, vol. 123. Amsterdam: IOS Press, 2005, p. 92.

42.

Grüninger

Fox

MS.

Methodology for the design and evaluation of ontologies. In: Workshop on Basic Ontological Issues in Know ledge Sharing, International Joint Conference on Articial Intelligence, Montreal, 1995.

43.

Beamon

Ware

TM.

A process quality model for the analysis, improvement and control of supply chain systems. Logist Inf Manag 1998; 11: 105–113.

An ontology-based approach for supply-chain quality control: From a principal–agent perspective

Abstract

Keywords

1. Introduction

2. Related works

2.1. Supply-chain quality control

2.2. Moral hazard in supply-chain quality control

3. Principal–agent mathematical models and data-centric framework

3.1. Principal–agent mathematical models

3.2. The data-centric framework

Definition 1

Definition 2

Definition 3

4. The ontology-based approach

4.1. The ontology-based models

4.1.1. Ontology metaclasses

4.1.2. Static, social, intentional and dynamic ontologies

Definition 4

Definition 5

4.1.3. Ontology development for supply-chain quality control

4.2. Data-centric algorithms

4.2.1. The artefact and service classes

4.2.2. Artefact classes

4.2.3. Service classes

4.2.4. Risky suppliers/product materials and losing efficacy contracts identification algorithms

Definition 6

Definition 7

Definition 8

5. Evaluation

5.1. Case study

5.1.1. Milk supply-chain quality-control description

5.1.2. Ontology development

5.1.3. Quality-risky suppliers and products and efficacy-losing contract identification

5.2. Comparison against traditional principal–agent contract-based approach using simulation data

5.2.1. Related mathematical model descriptions

5.2.2. Simulation

5.3. Comparison against other approaches in supply-chain quality control

6. Conclusion

Footnotes

Declaration of conflicting interests

Funding

References