This study is contributing to enhancement of safety standards, monitoring in industries by implementation of latest state of the art methods on rare-event classification of machine failure. Classification of machine failure is important to Maintaining and Monitoring the safety standard practices in industries. It also drives the cost and huge impact on industry performance, ratings, and reputation etc. Iterative machine failure can contribute to deviation from safety standards, this may lead to closure of industries. For any industry it costs a lot for machine failure using classic corrective maintenance with a cost of replace and shutdown of process. Hence this research contributes to reducing the cost of maintenance along with condition-based maintenance with high precision. The industries can take necessary steps to maintain the machines based on the prediction generated by the model. This model will also help in reducing the costs spent by Industries to target the machines with corrective maintenance. The model can be extended to study the real-time behavior of machines in different environments before they start to show signs of failure.
Machines and equipment play central role in organizations- needs, prosperity, and development. The cycle of development demands more from these machinery and equipment; hence, topic of machine failure and machine maintenance also appear to be important. In this study we determine the machine failure by construction of classifier model on predictive maintenance dataset. Taking the key factors affecting tool life as characteristic attributes. This study analyses the behavior of their effects on machine failure. Another purpose of the study is to solve the problem of rare-event class distribution in the dataset. Hence, developing a multitask learning based model to address both the issues. Multi-task auxiliary learning takes advantage of relevant auxiliary tasks to improve interpretation of a primary task. Due to unequal distribution, the generalization performance of classification model affects deeply. To get better performance from model in this study, I am using autoencoders for reconstruction of the input in AI4I 2020 Predictive Maintenance Dataset. Findings were examined by considering latest classification architectures and evaluation rubrics. It is expected to enhance the generalization performance of multitask model by addition of reconstruction error as auxiliary task. We then want to test the performance of autoencoder to the associated with standard neural network not in any way reduces performance and can enhance generalization. This study shall conclude the model performance for rare-event classification in different classifier settings and its implementation in machine failure for real time application.
The detection of failures and flaw in manufacturing tools and equipment has demonstrated, to be a challenge to scale its assurance and execution. Deflation in manufacturing tools and equipment takes place owing to many elements, generally tool wear, strain, heat failure and power failure. The aim of this study is to determine the machine failure by construction of classifier model on predictive maintenance dataset. The class imbalance data compromise the performance of the constructed model and this is addressed by assessing the oversampling methods with Multi-Task Learning (MTL)architecture. Also, to gauge the performance of auxiliary learning towards the advancement of the primary task learning.
Dataset Description- https://archive.ics.uci.edu/ml/datasets/AI4I+2020+Predictive+Maintenance+Dataset
Target variable Machine failure indicates machine failure at particular datapoint for any of following failure mode are true. Five independent failure modes of machine failure are as follow:
- tool wear failure (TWF)
- heat dissipation failure (HDF)
- power failure (PWF)
- overstrain failure (OSF)
- random failure (RNF)
If at least one of the above failure modes is true, the process fails and the 'machine failure' label is set to 1. It is therefore not transparent to the machine learning method, which of the failure modes has caused the process to fail. Considering timeline and resources of study, study is limited to machine failure only. Eliminating five failure mode from dataframe.