An investigation of deep learning approaches for efficient assembly component identification

Abstract

Background

Within the manufacturing sector, assembly processes relying on mechanical fasteners such as nuts, washers, and bolts hold critical importance. Presently, these fasteners undergo manual inspection or are identified by human operators, a practice susceptible to errors that can adversely affect product efficiency and safety. Given considerations such as time constraints, escalating facility and labor expenses, and the imperative of seamless integration, the integration of machine vision into assembly operations has become imperative.

Results

This study endeavors to construct a robust system grounded in deep learning algorithms to autonomously identify commonly used fasteners and delineate their attributes (e.g., thread type, head type) with acceptable precision. A dataset comprising 6084 images featuring 150 distinct fasteners across various classes was assembled. The dataset was partitioned into training, validation, and testing sets at a ratio of 7.5:2:0.5, respectively. Two prominent object detection algorithms, Mask-RCNN (regional-based convolutional neural network) and You Look Only Once-v5 (YOLO v5), were evaluated for efficiency and accuracy in fastener identification. The findings revealed that YOLO v5 surpassed Mask-RCNN in processing speed and attained an mean average precision (MAP) of 99%. Additionally, YOLO v5 showcased superior performance conducive to real-time deployment.

Conclusions

The development of a resilient system employing deep learning algorithms for fastener identification within assembly processes signifies a significant stride in manufacturing technology. This study underscores the efficacy of YOLO v5 in achieving exceptional accuracy and efficiency, thereby augmenting the automation and dependability of assembly operations in manufacturing environments. Such advancements hold promise for streamlining production processes, mitigating errors, and enhancing overall productivity in the manufacturing sector.