Summary
This project proposes a methodology that applies multi-threshold count feature extraction at multi-resolution level using wavelet packet transform, which will show a redundant and non-optimal feature map from the AE signal. Next, recursive feature elimination will be applied to reduce and optimize the huge number of predicting features generated in the data acquisition step, and random forests regression will be performed to provide the projected tool wear/condition. The methodology will be applied for selected drill bits under pre-established material selection and cutting conditions. To do the experiment, a high quality table top CNC drilling machine will be used and the AE sensor will be attached at the tool holder place of drill bit. The signal collection from AE sensor will be done using National Instrument’s data acquisition card based on virtual Instrumentation technology, high quality signal processing components and Lab VIEW software. In the next phase of the research work, the results obtained from the previous step will be taken to deep learning based analysis. There are various techniques involved in deep learning to analyse the data which is collected from the AE sensors. In particular, the results will be analyzed and compared with several Machine learning algorithms such as support vector machines, k-nearest neighbors and decision trees. The analysis using these deep learning algorithm is a latest trend in tool wear and condition monitoring and the research team believes that the experimental results will reduce the error in predicting the condition of the drill bit. The team is expecting the accuracy of the results in condition monitoring in drilling tool wear monitoring using deep learning techniques will be more than 97% which is the high level accuracy in health condition monitoring. The novelty of this work is the proposal of a novel approach for tool wear monitoring in drilling using a count based approach for feature extraction considering wavelet packet transform and using the so extracted features as input for random forest regression (RFR) and comparing its performance with other well used and researched ML techniques like ANN, SVM and decision trees.
Objectives
- To apply acoustic emission technology in drill tool condition monitoring to predict drill conditions using a count-based approach for feature extraction considering wavelet packet transform.
- To develop a data acquisition system using Lab VIEW software and virtual Instrumentation technologies.
- To analyze the extracted features and apply a suitable dimensionality reduction technique to identify suitable features, which are fed as input for random forest regression (RFR) and comparing its performance with other well used and researched ML techniques like ANN, SVM and decision trees.
- To compare the results of tool condition monitoring between deep and conventional machine learning techniques (ANN, SVM, KNN and DT)
- To implement and to prepare a documentation resources for drill tool monitoring system using modern technologies like Artificial intelligence, Machine learning, Cloud storage etc.
- To share the information to the manufacturing sectors in applying drill tool monitoring system.
Funding Agency
MOHERI
Collaboration
Collaborator 1: NMAM Institute of Technology, Nitte,India
Collaborator 2: Shiraz University of Technology, Iran