Multi-Modal Explainable Artificial Intelligence for neural network-based tool wear detection in machining
Document Type
Article
Publication Date
3-15-2025
Department
Department of Mechanical Engineering-Engineering Mechanics
Abstract
As data in manufacturing becomes more accessible and larger in scope, solutions are required to explain the underlying reasoning behind predictive models that use data from various modalities. This research aims to develop multi-modal Explainable Artificial Intelligence approaches to facilitate human-interpretable solutions for tool wear prediction in machining. Initially, a multi-modal neural network was used as the supervised Machine Learning classifier for training and binary classification by applying a set of features comprised of different modalities. The data consisted of two sets of image data representing the Flank and the Rake views of the tools accompanied by time series data including acceleration, acoustics, temperature, spindle speed, and feed rate during the orthogonal tube turning process. The classifier was used to predict the condition of the tool after the cutting process. After the training process and the network performance evaluation were completed, two multi-modal neural network explainability approaches were investigated. The Full Multi-Modal Explainable Artificial Intelligence approach provided a general explainability of the feature importance based on latent space representations of the input data while the Decomposed Multi-Modal Explainable Artificial Intelligence approach produced explainability results of input feature importance by decomposing the multi-modal model into its constituent single modal sub-models. The developed explainability methods were demonstrated to provide sufficient information regarding explaining the multi-modal network performance and the decision-making processes. Hence, the presented methods will enable end-users to understand the underlying logic for advanced neural network-based models that incorporate data from different modalities through the selection of multiple algorithms.
Publication Title
Engineering Applications of Artificial Intelligence
Recommended Citation
Sotubadi, S.,
Pallissery, S.,
&
Nguyen, V.
(2025).
Multi-Modal Explainable Artificial Intelligence for neural network-based tool wear detection in machining.
Engineering Applications of Artificial Intelligence,
144.
http://doi.org/10.1016/j.engappai.2025.110141
Retrieved from: https://digitalcommons.mtu.edu/michigantech-p2/1441
