Monitoring and controlling 3D printing process

点击翻译按钮获取中文摘要

Uncertain manufacturing processes, like 3D printing, additive manufacturing (AM) processes and micro/nano-machining processes, are sensitive to random variables, such as environmental factors (temperature, humidity, PM2.5 particles, vibrations), material density distribution, power supply stability, and machine degeneration/fatigue and errors. It is difficult to use an uncertain manufacturing process to effectively and economically mass produce products. Furthermore, it is ineffective to use a traditional deterministic closed loop control model to control an uncertain process since the control model keeps changing with the random variables. To solve the problem, this project will employ a Kriging model (a non-deterministic model subject to a confidence interval) to control a 3D printer. To detect errors, digital twins of the 3D printing process will be compared with on-line measuring quality data (printing layer geometric shape and surface finishing parameters) that has been gathered from a camera by using image processing and machine learning techniques. In terms of the errors, the Bayes model can statistically learn the random variables and update the confidence interval of the Kriging model. Consequently, the controller parameters are modified and adaptively adjusted to the random variables iteratively. The success of project may provide novel cloud monitoring and controlling services for 3D printing companies and leads to a new generation of AM technology, which is able to intelligently adapt to the random environmental, material and machine impacts or disturbances to keep the process stable. Obviously, this new generation of AM technology can be widely applied in various businesses to economically mass produce customized products. This will make Canada and Alberta leaders in economically competitive additive manufacturing, which creates great economic potentials not only for manufacturing companies but also for other businesses. To easily implement this new technology, we will also devise a soft sensor technology through applications of image processing and machine learning techniques to use a camera to measure the quality data and process faults, e.g., filament supply loss and nozzle clogging.