Ensemble Learning Approach for Predictive Maintenance in Investment Casting Process

Beyond the Automation Pyramid, industries are currently embracing intelligence. One of the challenges in Industry 4.0 is to conduct Predictive maintenance (PdM) for the Investment Casting Process, which is one of the oldest metal-forming industrial processes. According to the existing works, PdM is achieved by data-driven methods for scheduling just-in-time maintenance. However, traditional Machine Learning (ML) techniques can make good predictions to some extent but are not guaranteed to be accurate. This limitation served as the driving force behind the current research work, which examined ensemble technique-based algorithms and compared them with standalone ML algorithms for the investment casting dataset. The most effective technique is used to achieve an optimal outcome i.e. gradient Boosting, it predicted the possible proportion of alloys during fault occurrence with 83.63% accuracy and 5.505 Average MSE. As a result, industries can learn to identify multiple root causes of failure, such as possible environmental conditions or proportions of the Investment Casting Materials, and predicts the possibilities of defects such as Flash, Misrun, Ceramic Inclusion, Shrinkage, and Slag Inclusion to arise. It has been proved that the proposed approach can also prevent downtime and optimise the performance of just-in-time maintenance.