bestseller-oriented vehicle specifications by car maker

1. Some Remarks on Bayesian Networksin Automotive Industry Prof. Dr. Rudolf KruseUniversity of Magdeburg kruse@iws.cs.uni-magdeburg.deDagstuhl 2008

2. Example 1 (Volkswagen) STRATEGY OF VW GROUP prefer individual vehicle specifications by customers bestseller-oriented vehicle specifications by car maker Marketing strategy very large number of possible variants low number of possible variants Complexity Vehicle specification 2,8L 150 kW spark Item short back Type alpha leather, Type L3 ...... yes 4 body variant door layouts seat covering vanity mirror ...... Item family engine radio

3. „Golf“ Class of Vehicles • approximately 200 item families (variables) • from 2 to 50 items in each family • i.e. more than possible vehicle specifications • choice of valid specifications is restricted by RULE SYSTEMS (10.000technical rules, even more marketing-, and production-oriented) Example (technical rules that restrict validity of item combinations) if then if and then How to handle installation rates of item combinations?

4. Problem Representation Rules for the validity of item combinations (specified for a vehicle class and a planning interval) Sample of produced vehicle specifications (representative choice, context-dependent, f.e. Golf) System of rules Historical Data If engine = e1 and auxiliary heater = h2 then generator in {g3,g4,g5} ... (Golf, short back, 2.8 L 150 kW spark engine, radio alpha, ...) ... ? predicted / assigned planning data (production program, demands, installation rates, capacity restrictions, ... bills of material, ...) Prediction Planning

5. Planning Model based on Belief Change Historical data System of rules context-dependent sample of produced vehicle specifications context-dependent rules for the validity of item combinations context : vehicle class, planning interval Use cond. independencies (Composition) Estimate prior distribution of installation rates Quantitative Learning PGM (Markov network) having the structure of the relational network Modify representation Transformation into a relational network with hypertree structure Revision Adaptation of installation rates of item combinations that change from valid to invalid Updating Find referential for item combinations that change from invalid to valid Fusion fused consistent Markov network for item planning Planning Model

6. Network: VW Bora 186 variables 174 cliques max. 9 dimensions

7. Decomposable Models Graphical Model  Decomposition + Local Model + Operations 1. Decomposition 2. Local Models 3. Operations  Dir./ Undir. Graph,  Relational Constraints,  Belief Change: Hypergraph, Possibilistic Constraints, Conditioning, Clique Trees Probabilistic Constraints Revision, Updating

8. Operation Focussing (in real time due to decomposition) • Input Data : item combination (set of variable instantiations) • Operation : Calculate the conditioned network distribution and the probability of the given item combination (propagation). • Application : Calculation of part demands Compute the installation rate of item combination . Simulation Analyze customers‘ preferences with respect to those persons who buy a navigation system in a VW Polo.

9. Planning Operation Revision • Input Data : Family of marginal / conditional probability distributions • Operation : Calculate Markov network with same structure that satisfies all input distributions and is conform to the principle of minimal change. • Application : Marketing stipulations Installation probability of item air condition increases by 10 % in case of Golf all-wheel drive in France. Logistic restrictions The maximum availability of engine in week 32/08 is 1.000 .

10. Software in Daily Use Worldwide • Project Leader: Jörg Gebhardt • Client-Server System • Server on 6-8 Machines (16 GB each) • 4-Processor AMD Opteron system • Terabyte storage device • Linux, JAVA, Oracle • WebSphere Application Developer, Eclipse • 4000 differnet networks are in use VW-Motivation for using the Bayesian Network solution: Better Performance than old system, Best Solution (several proposals were examined by planners) 18

11. Example 2 (Daimler) Explorative Data Analysis • Intuitive Views on the Data • Application-driven Pattern Selection Extracting Patterns from Dependency Models (Bayesian Networks) Focus on Temporal Changes in Patterns • Visualization of Changes Identifying Interesting Patterns with Linguistic Descriptions

12. Bayesian Network Complete network after dependency analysis at Daimler research plant (global learning methods,...) Which attribute values, have what kind of impact on the failure? Are theretemporal dependencies?

13. Explorative Data Analysis Ratio between (marginal) Aircondition sale rate and sale rate given the Country Airconditions of type 1 fail much more often in Egypt and Oman Marginal Country distribution Relative frequencies of Engine given Country Example of a user interface

14. Visual Pattern Discovery Subnet of the entire network 900 vehicles 1200 vehicles Partition of the entire vehicle set according to the network’s potential tables. Interesting patterns (rules) are represented by circles

15. Temporal Changes • Patterns (describing failed vehicles) do not arise out of a sudden. • Rather: Evolvement as time progresses. • Management decisions: Again, it takes time to see an effect. • Therefore: Consider the temporal changes of pattern properties.

16. Linguistic Descriptions Problem: Pattern finding algorithms return large number of results. Assessing the patterns manually would be infeasible. Approach: Use linguistic concepts to reduce the number of retrieved patterns. Example: Return only rules with • approximately unchanged lift and • slightly increasing support

17. Example • Before: All rule trajectories over time.

18. Example • After: Rule trajectories over time meeting the linguistic concept.