Authors:

Vidit Hemal Patel

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

In recent years, advancements in Internet of Things (IoT) and Artificial Intelligence (AI) have significantly boosted interest in smart manufacturing. This has led to increased use of Computer Numerical Control (CNC) machines, which can be automatically controlled by a computer. However, automated monitoring of these machines remains a significant challenge. To address this issue, we propose a real-time monitoring system that analyzes the sound signals emitted by the machines using a Convolutional Neural Network (CNN). To test this hypothesis, we installed a sound sensor at the base of a 16-cylinder shaft grinding machine, capturing sound signals over a 90-minute period during which the machine completed seven different grinding cycles. Mel-spectrograms were generated for each cycle and analyzed. We labeled different segments of the spectrograms into distinct classes, such as Grinding, Running, Idle and Rapid-movement, creating a labeled dataset for training the CNN. We then trained three models: one classifying Grinding, Running, and Rapid- movement; another distinguishing between Grinding and Running; and a third classifying Grinding, Running, and Idle states. Due to the limited amount of data available for Rapid-movement class, the first model lacked accuracy, but the other two models accurately predicted the activities they were trained to recognize. These results demonstrate that sound analysis using a CNN has significant potential for automation in machine monitoring. Potential applications of this research include enhanced machine maintenance, early fault detection, and improved operational efficiency. However, to fully realize this potential, further data collection is necessary to train the CNN model to classify a broader range of activities accurately, and more extensive testing of the model is required. Keywords: Convolutional Neural Network; Data Analysis

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Vidit Hemal Patel