Measuring the impact of daily workload upon plant operator production performance using artificial neural networks

Previous research has identified that six factors (each of which comprises numerous variables) impact upon plant operator productivity. These factors are: daily workload, motivation, management, maintenance, education and training, and stress and fatigue. In order to explore and better understand the complex interrelationships that exist between these factors, an iterative empirical examination of each (in turn) was undertaken. This article reports on analysis of the first of these factors (daily workload), which comprises 51 variables identified from earlier pilot studies. A field study practical experiment and follow-on questionnaire were used as the data capture mechanisms. The experiment produced productivity time data, while the questionnaire yielded a mixture of interval (Likert scale) and Boolean data. The standard deviation technique was utilized to define and delineate the boundaries of the dependent variable (excavation time) into good, average and poor categories of plant operator production performance. A generalized feed-forward artificial neural network model was then created to simulate plant operators' productivity. Cumulatively, outputs derived from mean squared error, confusion matrix and correlation coefficients highlighted that the developed model is reasonably accurate, although further improvement to classification performance could be made. Finally, sensitivity analysis revealed the impact that each independent variable had upon predicting the dependent variable while also identifying which were non-significant in terms of being predictors. The latter exercise confirmed that 19 of the original 51 variables could satisfactorily categorize operator production performance.