Notifying Type-2 Error and Segregating Undefined Conditions in Health Monitoring of Milling Cutter: A Statistical and Deep Learning Approach

PurposeArtificial Intelligence-based methods have significantly assisted in condition monitoring of cutting tools by using supervised learning algorithms. When a machine learning model or a neural network is used for classification, it yields four different cases: true classifications (includes true positive and true negative), type I error, and type II error. True classifications encompass all data points that were accurately classified as healthy or damaged. The type I error refers to a true negative label misclassified as positive, and the type II error refers to a true positive label misclassified as negative. The type II error goes unnoticed when the model is deployed in real-time, while the other three cases are notified successfully. Notifying the type II error is the genuine concern, i.e. when a faulty label is misclassified as healthy.MethodsIn an attempt to address this problem, an enhanced neural network was constructed, and the values derived from the outputs of neurons are used to facilitate notification of type II error by controlling the confidence levels. Another problem arises while resorting to such learning algorithms: the inability to classify data into labels besides those used for training. This type of undefined data is encountered due to various unknown factors stemmed due to a high noise environment. To tackle this, a discriminator is used to evaluate the training data distribution with the incoming data and segregate undefined conditions into a separate undefined category.ResultsThe vibrations caused by unknown factors are fed into the network as an input, triggering the discriminator to classify these data points into a separate category. The proposed methodology aims to notify type II error caused by AI models in real-time and simultaneously also seeks to establish a distinction between data and noise by segregating data points derived from noise into a separate category in real-time.ConclusionThe proposed provision would allow for better resource management and enhancement of efficiency of the system by mitigating the chances of a misclassification by the AI model.