FORECASTING EXPECTED DEMAND FROM AUTOMOTIVE SERVICE SHOPS, USING SELF-ORGANIZING NEURAL NETWORKS "SOM", ON A PARALLEL COMPUTING PLATFORM "CUDA"

José Antonio Sánchez González, Viridiana Hernández-Herrera, Moisés Márquez-Olivera

Abstract: This paper presents an unsupervised neural network schematic, on a parallel computing platform, for the prediction of expected demand from automotive service shops. Demand prediction is done by applying artificial neural networks of self-organized maps "SOM" and to optimize computing resources, parallel processing "CUDA" is applied. From the clusters generated by the network, the common characteristics of the vehicles that present a specific failure are identified, among the characteristics are, the mileage, the time without a visit to the workshop, average visits to the workshop, vehicle model, year of the vehicle, average mileage between each visit, number of maintenance performed, number of reported failures and time of seniority as an after-sales customer. In the training of the network, data obtained from the extraction of information from 60 Hyundai brand dealerships was used. Subsequently, new experimental data were added to the network to validate the proposal. Finally, it is demonstrated that the use of self-organized neural networks manages to generate the "Clusters" that allow predicting the expected demand regarding the type of failures that will occur in the vehicles that will inform the service workshops. Likewise, the processing time was optimized through the use of the parallel computing platform C"UDA".

Keywords: Expected Demand, Fault Diagnosis, CUDA, Neural Networks, Self-Organizing Maps, Diagnostic Complexity, Productivity, Efficiency, Cluster

DOI: 10.24874/PES08.02A.020

Recieved:   Revised:   Accepted:   
UDC:

Reads: 0   

---