Session: 03-16-01: AI Integration in Mechanical Engineering and Smart Manufacturing

Paper Number: 144456

144456 - Ai-Based Machine Condition Monitoring System 

This paper describes the approach for an AI-based machine condition monitoring system that utilizes machine learning (ML) and deep learning (DL) techniques to monitor and analyze the condition of machines in various industries. This system employs advanced algorithms and ML models to continuously collect data from sensors, such as vibration sensors, level sensors, and vision sensors, among others. By analyzing this data, the AI system can detect anomalies, predict potential failures, and provide valuable insights into the health and performance of the machines.

To demonstrate the feasibility of the concept, the system has been created and implemented for a rotating bowl feeder, a common piece of machinery used for the orientation of parts. Parts ranging from small washers to long-length screws can be dropped into a bowl feeder to be oriented for the next process. As parts become larger and more complex, bowl feeders tend to have more difficulty feeding parts at a rate and can experience jams. This issue can be compounded if an operator is working on other machinery and doesn't notice a fault immediately. This paper focuses on using AI for bowl condition monitoring using an ultrasonic level sensor, vibration sensor, part counting sensor, and vision sensor.

The experimental setup for this research work uses a SICK MPB10 Vibration Sensor, IFM UGT594 Ultrasonic Distance Sensor, and IFM OPU204 Fork Sensor for data collection. Using this data a ML algorithm, it is possible to train the algorithm to classify different operating statuses, such as normal operation or jam. Alternatively, a vision-based DL system using the Cognex Insight 2800 vision sensor is also added to classify different operating conditions. These conditions can be used to create ground truth labeling, which can increase the confidence in the data for a ML algorithm. The control system setup uses an IAI RCP6 Linear actuator and CompactLogix 1769-L27ERM PLC to optimize settings for a specified part feed rate, prevent or correct jams automatically, and offer better notification to operators if a jam or fault occurs.

To achieve this goal, a screening Design of Experiment (DOE) is first performed to narrow down the number of factors that have a significant impact on the feed rate and jam rate of the bowl. From the results of the screening DOE, three factors are identified for a full factorial DOE. These factors are bowl level (fullness of the bowl), throttle gate position, and vibrating motor amplitude. Preliminary tests are then developed and performed to identify the high and low limits to be used for the design of the experiment. These preliminary tests highlight operating parameters of interest to further investigate and narrow down the ranges for the design of the experiment, reducing the number of tests needed to be performed and saving time. The full factorial DOE will map out the interactions between the factors and what combinations will result in optimal feed rates and lower jamming rates.

During preliminary testing, sensors mounted to and around the bowl record vibration, feed rate, and bowl level, along with acquiring images to label the data and train a machine learning algorithm. The ML algorithm is then tested for accuracy during the full factorial DOE and compared to the DL algorithm developed on the Cognex Camera. The accuracy and classification results of the DL-based vision system will be reported as a confusion matrix, and the sensor data-based ML system performance will be reported as precision, recall, accuracy, and F1-score.

With the AI algorithms and the results of the DOE, we will have the ability to detect and quantify bowl conditions and understand what adjustments need to be made to the factors. With this information, a control loop can be developed to continuously monitor and adjust the bowl feeder to optimize the feed rate, prevent jamming, and better notify operators of the bowl's condition. The approach can be generalized and applied to other similar rotating machinery.

Presenting Author: Cameron Morris Western New England University

Presenting Author Biography: Cameron is a graduate student in the department of mechanical engineering at WNE. His area of research is AI, robotics, and machine vision.

Authors:

Cameron Morris Western New England University
Vedang Chauhan Western New England University
Sunny Panchal Scaler Neovarsity, Woolf University, Bangalore, India