Prediction of maintenance workforce efficiency using neural networks, fuzzy inference system and autoregressive fractionally integrated moving average for a process industry

Abstract

This study establishes the efficiency of the maintenance workforce in a process plant, utilising combined models, including artificial neural networks (ANN)-weighted aggregated sum product assessment (WASPAS) and ANNfuzzy inference system (FIS)-WASPAS. ANN models with 12 architectures and a maximum epochs ranging from 64 to 173 were the limits of the analysis. Real-life data were used to train, test and rank the different architectures. WASPAS was superimposed on the ANN model for selection and ranking purposes, while FIS rules were introduced to the ANN results to uncover the uncertainty and imprecision in the model. These results were compared with those of the autoregressive fractionally integrated moving average (ARFIMA) method. At different λ values, the 4-8-10-1 ANN architecture and the FIS with 18 rules were the fittest for the ANN-WASPAS and the ANN-FIS-WASPAS methods, respectively. The ARFIMA method revealed a coefficient of maintenance workforce efficiency of -1.8636x10-2. The constant parameter in the ARFIMA model and the root mean square error (RMSE) for the training and testing datasets are 0.9267, 0.0162, and 0.0150, respectively. In conclusion, the 4-8-10-1 ANN architecture more suitably predicts maintenance workforce efficiency than the 18-rule and ARFIMA models. This study predicts maintenance workforce efficiency, providing significant insights into the application of ANN, FIS, WASPAS, and ARFIMA models in a process industry.