Knowledge Capturing and Reusing in Engineering Change Management

Semantic Web and Case-Based Case Based Reasoning for Engineering Change Case Retrieval Change Management Hong Joo Lee November 12, 2004 Engineering Changes? Engineering Changes (ECs) are re-design or modification of products and components after product designs are released Introduction • Engineering Changes (ECs) have negative effects on development costs and time • However, ECs also have a role in improving the product quality & reducing unit cost • In order to reduce development costs and time, efficient management of EC processes and knowledge are important. Engineering Change Management (ECM) Problem Detection EC Request Form Identifying problem scope EC Evaluation Form Generating Alternatives Reference Reference EC Log Approval process EC Notice Form Limitations in ECM Research Problems • Collaborative experience through ECs can be easily lost after Engineering Change Orders (ECOs) are released. • Their experience and knowledge are not reused much at next similar situations. Research Objectives Tying EC processes into Knowledge Management Tying EC processes into Knowledge Management Providing a collaborative environment for accumulating ••Providing a collaborative environment for accumulating of EC knowledge and organizing of the various types of EC knowledge and organizing of the various types of knowledge of knowledge Suggesting a Knowledge Representation and Retrieval method Suggesting a Knowledge Representation and Retrieval method Case Study Case Study Overview • Case study of a new automobile development project at a major Korean automobile company – 10 semi-structured interviews – A collection of articles on engineering changes and new product development – Engineering Change Data from internal web based system EC Process of the host company Detecting Problems Manufacturing Team Problem Detect (Manufacturing) Cost Analysis Cost Management Y Cost Change? Discussing Problems Manufacturing Team Engineering Change Request Manufacturing Team ECR form Status Monitoring Rejected Rejected Engineering Change and Investigate Other Effects Engineering Not approved Analysis ECR -Definition of the change scope - Identify of the problem cause Review ECR and Assign Responding Engineer Engineering N ECO form Engineering Problem Solving Process (Engineering) Not approved Engineering approval Engineering Team ECO approval process (Engineering & Administrative team) Validate new data and distribute EO Technology Management Team PDM BOM Problems in ECM • Current system and the KM practices are failing to capture the various types of knowledge • Knowledge from an EC is distributed over diverse knowledge repositories • Retrieval of relevant knowledge through simple keyword-based search has some limitations Approaches for designing artifacts Approach Overview Collaborative Environment Collaborative Environment for ECM for ECM Collaboration Collaboration Model Model Domain Domain Ontology Ontology Semantic annotation Knowledge Retrieval and reusing Knowledge Repository Knowledge Repository Design Principles Perspective of Engineering Changes Characteristics of Engineering Change Process -Ad hoc Collaboration - Cross Functional - Distributed Environment - Knowledge Intensive Perspective of Collaborative Environment Organizational Perspective -Engineering Change Lifecycle -Project based roles Activity Perspective -Activity Management -Coordination Context Perspective -Process, Activity -Person Collaboration Collaboration Model for Model for ECM ECM Collaboration Model 1..* Ontology 1..* Knowledge Context 1..* 1 1..* Role 1..* 1..* 1..* Actor 1..* Milestone 1 1 1..* 1 1..* Discussion 1 1 is_super_of 0..* 1..* 1..* 1..* Knowledge Item 1..* Process Activity 1..* Document 1 1..* Conversation 1 Event Schedule Output document 1 1..* 1..* ECR ECO Domain Ontology – Automobile Development • Ontology – Each ontology Oi is defined as a tree of concept nodes, C ki (k = 1, 2, …) – Characterize Knowledge items and ECs – For automobile development, • • • • • Component ontology Problem ontology Process ontology Product ontology Solution ontology Ontology Examples Product ontology Vehicle Solution ontology Solution Car Bus Truck Stamping Welding Assembly Passenger Car SUV Van Part Function Form Compact Midsize Full size Luxury Hole Relocated Sizing Deleted Knowledge Representation and Retrieval Annotation Procedures Knowledge Annotation RDFS RDFS Collaboration Collaboration Model Model RDF RDF Schema RDF RDF Domain Domain Ontology Ontology Characterize Knowledge Knowledge Item Item Knowledge Knowledge Context Context Knowledge Retrieval • Query based Retrieval – RDF Query Languages are proposed in many studies such as RQL • Case Based Reasoning Approach – Consider diverse factors of ECs to calculated similarities – Applying pertinent information from the old case to the new situation Similarity Measure for complex products? Definitions of Similarity Measure • an EC case – An EC case is Ej = {akj | akj ∈ Opd ∪ Opb ∪ Opc ∪ Oc ∪ Os and k = 1, 2, …, number of concepts in the case}. • Similarity between two concepts in a single ontology sim(c pi , cqi ) = - log N ({E j | cri ∈ E j }) N (U ) Adopted from Resnik(1999) • Similarities between two EC cases w ∑ × f × sim (c Sim( E , E ) = w ∑ 1 2 i i i i i i 1i , c2 i ) CBR: Similarity Measure Problem p=1 info = 0 p (c ) = Information Value freq ( c ) N Assembly p=0.3 info = 1.2039 = - log p (c) p=0.1 info = 2.3025 Fixing Installation Resistance Implementation p=0.12 info = 2.1202 p=0.05 info = 2.9957 p=0.03 info = 3.5065 Screwing Clipping Stapling Coupling Fitting p=0.01 p=0.03 p=0.06 info = 4.6051 info = 3.5065 info = 2.8134 p=0.001 info = 6.9077 p=0.049 info = 3.0159 Selecting Weights of ontologies • Using the Analytic Hierarchy Process – A set of pair-wise comparison data of the ontologies was collected from the experts in the target company Solution Product Process Problem Component 0 0.1 0.2 0.035 0.094 0.214 0.21 0.447 0.3 0.4 0.5 Normalized Weights An Example of Similarity Calculation .. X8 L2.0 2003 Europe FrontSuspension F301 Screwing Pilot production … .. X7 F2.0 2000 Asia FrontSuspension F102 Fitting Pilot production

Problem ontology Problem Component ontology p=1 info = 0 Engine p=0.2 info = 1.6094 p=0.3 info = 1.2039 Axle Component p=1 info = 0 p=0.3 p=0.2 Performance Assembly info = 1.2039 info = 1.6094 Climate Chassis p=0.4 Control info = 0.9162 p=0.1 info = 2.3025 Steering Pedal Fixing p=0.1 info = 2.3025 Installation Resistance Implementation Axle p=0.2 Suspension Mechanism info = 1.6094 Fitting Front Rear Suspension Suspension p=0.05 p=0.1 info = 2.9957 info = 2.3025 Brake p=0.05 info = 2.9957 Coupling Screwing Clipping Stapling p=0.03 info = 3.5065 p=0.001 p=0.049 info = 6.9077 = 3.0159 info p=1 info = 0 Product ontology Bus Car Vehicle Process ontology p=1 Process info = 0 Engineering Prototype Pilot Manufacturing Truck p=0.7 info = 0.3566 p=0.5 Passenger Car info = 0.6931 SUV Van Prototype #1 p=0.2 info = 1.6094 Prototype #2 Assembly PlanningPilot Production p=0.05 p=0.1 info = 2.9957 info = 2.3025 Compact Midsize Fullsize Luxury p=0.1 info = 2.3025 X8 A3 M4 X5 X6 X7 p=0.05 p=0.03 info = 2.9957info = 3.5065 An Example of Similarity Calculation Similarity between concepts Factor Same Class No Yes No Yes Subsumed Concepts Assembly, Problem Front Suspension Luxury, Passenger Car, Car, Vehicle Pilot production Maximum Similarity Value 1.2039 2.9957 2.3025 2.3025 fi wi, weight of ith ontolog y 0.21 0.447 0.094 0.035 0.214 Problem Compone nt Product Process Solution 0.5118 0.3358 0.9098 0.9678 - Similarity = 0.21 × 0.5118 × 1.2039 + 0.447 × 0.3358 × 2.9957 + 0.094 × 0.9098 × 2.3025 + 0.035 × 0.9678 × 2.3025 = 1.0864 0.21 + 0.447 + 0.094 + 0.035 Advantages of concept based CBR • Consider diverse factors • Similarity measure can be maintained and updated easily • Able to expand relevant knowledge retrieval by adding document or activity type ontology Performance Evaluation Experiments • Experiment Setting – Actual Engineering Change Case: 261 cases – Similar 5 cases are retrieved for every case using following methods • Comparison Methods – Component based retrieval (used by the host company) – Ontology based retrieval (suggested by this paper) – Keyword based retrieval • Keywords are extracted from engineering change cases through KLT-v200 • Feature selection (TF-IDF) Results Total Number of Similar Cases Component (1) Same Product Same Process Same Problem Same Component Average Similarity Duncan Test 1204 1198 668 211 1204 0.2281 Ontology (2) 1305 1289 1003 697 934 0.271 (2) > (1) > (3) Keyword (3) 1305 1268 838 616 202 0.1014 Discussion and Conclusion Comparison with other systems Huang et al., 2001 ECM Approach EC status tracking Collaborati on Support Approval process support Collaborative environment Organizational information EC forms Knowledge Capturing Related documents Context information Query matching Knowledge Retrieval Browsing past cases Case matching Workflow and form driven ECM O O EC request, evaluation, notice and log form - SAP ECM Workflow driven ECM O O O ECR, ECO The case study Workflow driven ECM O O O ECR, ECO CECM Activity and collaboration focus O O EC process life cycle management O ECR, ECO and their evolution history Associated documents through activities, conversation, and discussions Knowledge context and background information such as activity, actor, conversation and process EC reference number, Keyword Product, Process, Problem, Solution, and Component Concept-based CBR mechanism Attached documents Attached document EC reference number, Keyword Whole EC logs - Documentation of Change Process EC reference number, Keyword Product and parts - EC reference number, Keyword Products and parts - Benefits and Costs • Benefits – A rich set of knowledge items can be accumulated – Represent EC knowledge linked with contexts – Using the CBR technique for providing relevant past ECs • Costs – Initially require iterative processes for building ontologies – Require more efforts of engineers and manufacturers for using a collaborative environment Conclusion • Showed why accumulation and reuse of knowledge are crucial in ECM • The architecture for Collaborative ECM was designed • Presented a means to efficiently represent and retrieve EC knowledge Further Research • Empirical evaluation of the CECM for practical usefulness Question? CECM – A prototype system Architecture of CECM CECM Process support system Activity Knowledge Repository Knowledge Query Conversation support Approval Routing Actor Document Discussion Milestone Ontology Knowledge Delivery Ontology Management Ontology Knowledge Indexing Activity Management Raising ECR Case Retrieval Knowledge Context Evolution of the ECR