Case: predicting parts consumption
This case outlines how BayMiner is applied to the task of parts demand prediction in production. The case considers an application where capital goods are produced based on customer requests rather than production in series.
The customer faced huge cost pressure because their purchasing volumes were significantly lower than their competitors. The customer focused on certain narrow niches but their manufacturing process was not capable to handle the complexity this type of process required. This caused delivery delays and some regular clients had already switched supplier. Furthermore, the customer used a number of small subcontractors that could not deliver with very short delivery times. The situation became critical when the market turned unstable due to unpredictable actions carried out by competitors trying to win market share.
Overview - why probabilities
Bayesian networks can be used to calculate the probability that a piece of machinery to be produced will have a certain configuration. Depending on the configuration of the machinery a certain amount of parts of different types have to be used for the construction of it. The expected demand of a specific part can be computed based on the probability distribution of different machinery configurations in the past.
The customer has a number of options available when buying the machinery. The production of the machinery is constrained by market bound constraints such as type of use and legislation. These choices are specified by the manufacturer. The specification of the machinery is also determined from another ser of choices made by the customer. The choice of the customer with respect to each option is free under certain constraints but in reality patterns exist. The role of the Bayesian Networks is to identify these patterns.
Use of Historical Data
Today the customer records the past production in detail in a database. The database includes the information of a complete configuration of each machinery produced in the past. This historical data reflects both the technical and marketing constraints at the time of production, as well as the preferences of the customer. To achieve the maximum benefit of the knowledge model, the data that is the source to the model should include also the quantitative information to enable the model to identify strengths of the dependence relations.
Use of the Model
Once the model has been constructed from the historical data, the model is used to predict the demand for parts. The prediction of parts demand is based on the production of a number of machines in a past period. The BayMiner model is a representation of the average machine built during this period. To achieve a prediction for the next period, human intervention is needed to convert the predictions to suit the forecasted production during the next time period. The number of parts required for the production depends on the number of items to be produced, the configurations of the machinery and the number of parts used for each particular configuration. From these the predicted number of parts of a partial kind to be used in the future production can be estimated.
Update of Model
There is one major hurdle though. The items produced in the past have been constructed under a set of constraints, which may be different from the constraints, which have to be satisfied in the future. Therefore the rapid updating of the knowledge model is crucial. A revision of the BayMiner knowledge model can be performed quickly by just adding more cases (deliveries) to the table from which the original model was produced.
The customer could reduce their inventory by10 %, which led to significant cost savings and shorter production times. The intuitive nature of the BayMiner user interface also helped to generate explanations of the reasoning and the predictions performed by the model. The customer used these characteristics to validate the predications made by the model.