August 2023 Volume 5

FORGING RESEARCH

Forging Defect Detection Using Artificial Neural Networks Joseph Domblesky, Richard Povinelli, Sida Zhang, Ross Crowley,

and Philip Voglewede Marquette University

Introduction The North American forging industry’s use of digital-based technology increased tremendously during the 3rd Industrial Revolution (circa 1980s- early 2000s) and this has had a profound effect on best practices, particularly in engineering, administrative, and other support roles. However, technological advances from the ongoing 4th Industrial Revolution (a.k.a. Industry 4.0., circa 2010-present) continue to emerge at a rapid pace and can be expected to further change the way the forging industry operates. Some indication of the impact that Industry 4.0 will have can be found by noting that 70% of respondents in a recent poll taken by McKinsey & Company reported that this was one of their top strategic imperatives [1]. It is also worthwhile to note that Industry 4.0’s impact on manufacturing will be centered around the development of “smart” production systems as opposed to stand-alone islands of automation (e.g., robots, PLCs, etc.) that characterized Industry 3.0. As such, smart automation will be comprised of cyber physical systems where machines, robots, sensory feedback, and internet-based communications are supervised and controlled by machine learning algorithms such as Artificial Neural Networks (ANN). Consequently, Industry 4.0 based technology will have the most impact at the shop floor level in the form of autonomous manufacturing systems that incorporate Machine Learning (ML). While the forging process has traditionally been considered to be problematic with respect to implementing automation, one only needs to look at Artificial Intelligence (AI) based systems such as ChatGPT [2] and autonomous vehicle capabilities to appreciate how fast this technology has developed and recognize the benefits it will have for the forging industry. A review of the open literature shows that the current generation of ANN-based algorithms has advanced to the point where they can independently “learn” how to sort through extraneous inputs and variable data with minimal effect on accuracy. On a related note, it should also be noted that ANN-based systems now have the capability to detect and adapt to the typical levels of variation encountered in many manufacturing operations. Consequently, it must be considered that with appropriate refinements, smart automation systems can be developed to operate effectively in a forging environment. While forging automation is being implemented in a number of facilities, most efforts are at the Industry 3.0 level with relatively few Industry 4.0 applications in forging production at the present

time. This disparity can be attributed to several reasons. The first is that most forging companies are by necessity operating with lean workforces and are reluctant to devote scarce resources to research and development (R&D) without an immediate payback. Compounding this difficulty is that R&D efforts are also constrained because equipment is fully utilized in production work and tight delivery schedules preclude diverting plant resources. Traditionally the industry has relied on system integrators to develop hardware solutions and implement new technology. However, the technology associated with Industry 4.0 and smart automation is heavily reliant on application specific software and algorithms. Furthermore, as is the case with any new technology, there will be a significant learning curve and the workforce may require a significant amount of training to develop the requisite skill sets needed to operate and maintain the technology in the plant. However, given the world wide development of Industry 4.0 technology, the forging industry needs to be proactive in preparing for and adopting it in the near term. Given the above constraints, this begs the question of how smart automation and related technology could be implemented in the North American forging industry. One possible solution can be found in the European Union where collaborative partnerships between industry and university researchers, particularly in Germany, have compiled a well-established track record with respect to introducing advanced metalworking technology in manufacturing operations. One of the prerequisites for any new forging technology is to demonstrate its capabilities and identify specific challenges before full-scale deployment can be considered. A risk management plan typically includes building a prototype set-up to test the robustness and accuracy of the technology under forging conditions. To address this need, a project has been ongoing at Marquette University (MU) to develop smart forging technology based on a sensorized system, artificial neural networks, and data analytics. This article is the second in a 2-part series which describes R&D work being conducted at Marquette University’s Materials Processing Lab to develop and demonstrate smart forging technology. The first article discussed basic concepts related to ANNs and showed results from an open die prototype system. In the present article, prototype ANN systems were developed to perform defect prediction in an aluminum closed die forging and also to assess scale removal effectiveness in a steel buster operation.

FIA MAGAZINE | AUGUST 2023 67