Abstract:
In a competitive market, the desire for companies to provide quality products and/or uninterrupted services to satisfy customers and stay afloat in business is on ever increase. However, machine failure often leads to unprecedented downtime, impeding the production process and affecting businesses adversely. One big challenge companies face is to optimally utilize machine's remaining useful life while at the same time reducing downtime to “acceptable low” duration.Technological advancements have been leveraged in many ways by manufacturing industries, one of the ways being the use of sensors to capture large volumes of data representing the health state
of manufacturing machines/components. The insight embodied in collected data guides the decision process for system maintenance. A prominent analysis approach that has become increasing reliable is using machine learning to mine insight from data and use it to support decision making.
In this research, we train machine learning models in python using labeled time-series data collected on production machinery, and use the trained models to predict possible machine failure the next day, thus reducing the risk exposure of employees and improving the manufacturing process. By testing our models on validation dataset, the Multilayer Perceptron neural network reliably out-performed the other models with an accuracy score of 99.99994%. This model if validated with recent data and deployed would provide inside on system failure before it happens.
As a result, this will lower the cost of maintenance as planning can be done with relative ease,increase system availability since system failure becomes predictable and other measures can be taken before failure occurs, and ensure reduced customer dissatisfaction that results from long system downtime.
