Multi-Segmentation Parallel CNN Model for Estimating Assembly Torque Using Surface Electromyography Signals.

Area of Science:

• Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate bolt tightening torque is crucial for product assembly quality.
• Wrenches lacking torque measurement are common, necessitating effective monitoring solutions.
• Surface electromyography (sEMG) offers a potential, yet underexplored, avenue for torque estimation.

Purpose of the Study:

• To propose and evaluate a multi-segmentation parallel convolution neural network (MSP-CNN) for estimating assembly torque via sEMG signals.
• To enhance the accuracy and reliability of wrench torque monitoring in assembly processes.
• To investigate the impact of signal preprocessing and model parameters on torque estimation performance.

Main Methods:

• Development of a bolt tightening test bench to collect synchronized sEMG and torque data.
• Preprocessing of sEMG and torque signals, followed by segmentation into labeled torque ranges.
• Implementation and training of the proposed MSP-CNN model, comparing it with single CNN models and analyzing different pooling strategies.

Main Results:

• Average pooling demonstrated superior performance over maximum pooling in CNN-based torque classification.
• The MSP-CNN model achieved higher accuracy in torque monitoring compared to individual CNN models.
• The proposed model effectively addressed issues of non-convergence and slow convergence observed in standalone CNNs.

Conclusions:

• The MSP-CNN model provides a robust and accurate method for estimating assembly torque using sEMG signals.
• sEMG-based torque monitoring is a viable solution for improving quality control in mechanical assembly.
• The study highlights the potential of deep learning techniques in non-invasive industrial monitoring applications.