Capturing and reusing knowledge in engineering change management: A case of automobile development

Abstract

The development of complex products, such as automobiles, involves engineering changes that frequently require redesigning or altering the products. Although it has been found that efficient management of knowledge and collaboration in engineering changes is crucial for the success of new product development, extant systems for engineering changes focus mainly on storing documents related to the engineering changes or simply automating the approval processes, while the knowledge that is generated from collaboration and decision-making processes may not be captured and managed easily. This consequently limits the use of the systems by the participants in engineering change processes. This paper describes a model for knowledge management and collaboration in engineering change processes, and based on the model, builds a prototype system that demonstrates the model’s strengths. We studied a major Korean automobile company to analyze the automobile industry’s unique requirements regarding engineering changes. We also developed domain ontologies from the case to facilitate knowledge sharing in the design process. For achieving efficient retrieval and reuse of past engineering changes, we used a case-based reasoning (CBR) with a concept-based similarity measure.

Appendices

Appendix A. Brief introduction to the AHP method used to calculate the weights of the ontologies

We have O ₁, O ₂, ..., O _n as the criterion for which weight values should be calculated. Using the n criterion, an (n × n) matrix A can be defined where element a _ij of A is the experts’ preference of O _i over O _j when retrieving relevant engineering change cases. It is also assumed that the elements of A adhere to the following rules:

• Rule 1. If a _ij  = α, then a _ji  = 1/α, α ≠ 0.

• Rule 2. If O _i is judged to be of equal relative importance as O _j , then a _ij  = 1, a _ji  = 1; in particular, a _ii  = 1 for all i.

The final objective is to calculate the weight value w _i for each ontology O _i . Using the above matrix, w _i can be calculated using the following formula:

$$ w_{i} = \frac{1} {{\lambda _{{\max }} }}{\sum\limits_{j = 1}^n {a_{{ij}} w_{j} } }\,{\left( {i = 1,2,...,n} \right)}, $$

where λ_max is the maximum eigenvalue of matrix A (Saaty, 1980).

A set of pair-wise comparisons of ontologies was collected from the experts in the target company using questionnaires. Matrix A was obtained by integrating the experts’ preferences as follows:

	Component	Problem	Process	Product	Solution
Component	1	5.593	6.804	3.476	1.197
Problem	0.179	1	6.082	2.924	1.71
Process	0.147	0.164	1	0.281	0.147
Product	0.288	0.342	3.557	1	0.43
Solution	0.836	0.585	6.804	2.327	1

Consequently, we obtained the normalized weights of the ontologies:

$$ {\left\{ {w_{c} ,w_{{pb}} ,w_{{pc}} ,w_{{pd}} ,w_{s} } \right\}} = {\left\{ {0.447,0.21,0.035,0.094,0.214} \right\}} $$

The terms w _c , w _pb , w _pc , w _pd , and w _s are the weights of the component, problem, process, product, and solution ontologies, respectively.

Appendix B. Definition of the similarity measure

1.1 B.1 Definition of ontologies

Each ontology O _i is defined as a tree of concept nodes, C _ki (k = 1, 2, ...). (Note that when designating a specific ontology, we use the terms O _pd , O _pb , O _pc , O _c , and O _s for the product, problem, process, component, and solution ontologies, respectively.)

1.2 B.2 Definition of an engineering change case

An engineering change case E _j is defined as the set of concepts that correspond to the concepts in the ontologies, i.e., $E_{j} = {\left\{ {a_{{kj}} \left| {a_{{kj}} } \right. \in O_{{pd}} \cup O_{{pb}} \cup O_{{pc}} \cup O_{c} \cup O_{s} \;{\text{and }}k\; = 1,\,\,2,...,\;{\text{number of concepts in the case}}} \right\}}$.

1.3 B.3 Definition of the similarity between two concepts in a single ontology

The similarity between two concepts c _pi and c _qi in O _i is defined as (Resnik, 1999)

$$ sim{\left( {c_{{pi}} ,c_{{qi}} } \right)} = - \log \frac{{N{\left( {{\left\{ {E_{j} \left| {c_{{ri}} \in E_{j} } \right.} \right\}}} \right)}}} {{N{\left( U \right)}}}, $$

where c _ri is the closest common parent (CCP) to both c _pi and c _qi and U is the entire set of engineering change cases in the case base. N({E _j |c _ri  ∈ E _j }) is the number of engineering change cases that belong to concept c _ri . The CCP represents a node in O _I , which is the parent or ancestor of both of the two given concepts located closest to it. (Note that the numerator in the log term can become much smaller in large ontologies with greater depth and more concepts. For this reason, it is expected that the average value of the similarity is proportional to the size of an ontology.) For example, if we calculate a similarity value between the ‘Screwing’ and ‘Fitting’ concepts in the problem ontology in Fig. 6, the parents of the ‘Screwing’ concept are ‘Fixing,’ ‘Assembly,’ and ‘Problem,’ and the parents of the ‘Fitting’ concept are ‘Installation,’ ‘Assembly,’ and ‘Problem.’ Therefore, the common parents in this case are ‘Assembly’ and ‘Problem’ and the CCP is ‘Assembly’ because ‘Assembly’ is closer to ‘Screwing’ and ‘Fitting’ than ‘Problem.’

1.4 B.4 Definition of the compensation factor for reducing size effects (Inverse Concept Frequency)

The compensation factor f _i for O _i is defined as

$$ f_{i} = \log \frac{{{\sum\limits_m {N{\left( {O_{m} } \right)}} }}} {{N{\left( {O_{i} } \right)}}}. $$

(Note that this measure is similar to the inverse document frequency (IDF) measure widely used in information retrieval [Salton & McGill, 1983].)

1.5 B.5 Definition of the similarity between two engineering change cases

The similarity between two cases, E ₁ and E ₂, is defined as

$$ Sim{\left( {E_{1} ,E_{2} } \right)} = \frac{{{\sum\limits_i {w_{i} \times f_{i} \times sim_{i} {\left( {c_{{1i}} ,c_{{2i}} } \right)}} }}} {{{\sum\limits_i {w_{i} } }}}, $$

where c _1i and c _2i are the concepts of E ₁ and E ₂ , respectively, defined in the O _i dimension, c ₁ ∈ E ₁ , c _2i  ∈ E ₂ , c _1i  ∈ O _i , and c _2i  ∈ O _i .