1. First Steps Through Intelligent Grinding Using Machine Learning via Integrated Acoustic Emission Sensors
- Author
-
Mohammadali Kadivar, Bahman Azarhoushang, and Siamak Mirifar
- Subjects
0209 industrial biotechnology ,business.product_category ,Computer science ,Mechanical engineering ,02 engineering and technology ,Industrial and Manufacturing Engineering ,020901 industrial engineering & automation ,0203 mechanical engineering ,Machining ,Surface roughness ,lcsh:T58.7-58.8 ,Artificial neural network ,Mechanical Engineering ,online monitoring ,process prediction ,grinding ,Backpropagation ,Machine tool ,Grinding ,020303 mechanical engineering & transports ,Acoustic emission ,Mechanics of Materials ,Feedforward neural network ,business ,acoustic emission ,lcsh:Production capacity. Manufacturing capacity ,artificial neural networks - Abstract
The surface roughness of the ground parts is an essential factor in the assessment of the grinding process, and a crucial criterion in choosing the dressing and grinding tools and parameters. Additionally, the surface roughness directly influences the functionality of the workpiece. The application of artificial intelligence in the prediction of complex results of machining processes, such as surface roughness and cutting forces has increasingly become popular. This paper deals with the design of the appropriate artificial neural network for the prediction of the ground surface roughness and grinding forces, through an individual integrated acoustic emission (AE) sensor in the machine tool. Two models were trained and tested. Once using only the grinding parameters, and another with both acoustic emission signals and grinding parameters as input data. The recorded AE-signal was pre-processed, amplified and denoised. The feedforward neural network was chosen for the modeling with Bayesian backpropagation, and the model was tested by various experiments with different grinding and neural network parameters. It was found that the predictions presented by the achieved network parameters model agreed well with the experimental results with a superb accuracy of 99 percent. The results also showed that the AE signals act as an additional input parameter in addition to the grinding parameters, and could significantly increase the efficiency of the neural network in predicting the grinding forces and the surface roughness.
- Published
- 2020