This is a predictive maintenance model that predicts the machine break down based on sensor data.
A major problem faced in manufacturing or mining industries is significant costs which are associated with the down time of the machinery. So most of these businesses are interested in predicting the problems, which are responsible for down time. Predictive maintenance problem can be vary in different ways like failure prediction, failure detection and predicting the remaining life time of a machine. This model mainly focuses on when in-service machine will likely to fail, so that future maintenance can be planned accordingly. Not only the former one, but this model can also predict when the machine will be likely to fail. The business problem in our example is to predict possibility of a machine to fail within a window period. This model is a classification approach in predictive maintenance problem. **Input data:** The input data uses NASA dataset and you can download it from here. The dataset contains more than 160,000 observations and 26 columns in the training set and more than 106,000 observations in test set. Noise in the distribution of data may accounts in reducing the goodness of model, so we need to take care of it. Rolling mean and rolling sd comes to recue to deal with this problem. **Advantages of using this model** Reduces equipment costs, labour costs and down times due to machine failure. It increases the safety, revenue and efficiency of employee time. **Output:** There are two labels in this classification problem, one is label1 that predicts if the machine will break down within 10 cycles or not. Another output in this model is label2 which has been break down into 3 categories, 1. Machine that may breaks within 15 cycles, 2. Machine that may breaks in between 15 and 30 cycles, 3. Machine that will not break down.