November 2022 Volume 4

FORGING RESEARCH

Artificial Neural Network (ANN) and Multiple Input Multiple Output (MIMO) Systems to Monitor & Predict Forged Part Attributes – Part I: Bust By Joseph Domblesky, Richard J. Povinelli, Ross Crowley, and Philip Voglewede Marquette University, College of Engineering

aggregator that compiles individual part information to generate a digital twin of the forging process. To explore how an ANN/MIMO system could be implemented in a forging environment and ultimately in a production process, a multi-year project is being conducted at Marquette University. The project involves an interdisciplinary team of researchers from Mechanical and Computer Engineering who are participating in a collaborative 3-year project. The work involves a 3-phase effort that includes a) prototype demonstration and feasibility, b) scaled-up prototype process, and c) full scale deployment using both open and closed die processes. This article describes the results obtained from the Phase 1 effort focusing on open die forging. Similar work related to closed die forging and efforts to predict defects in an aluminum The basis of an ANN is that it is a network that mimics brain activity such that predictions can be made without building a mathematical model from first principles. One advantage of ANNs is that the inputs can consist of discrete or time-series data or a combination of both. Consequently, an ANN can rapidly process a large volume of data that involves complex, non-linear relationships and predict the likely outcome. In simplest terms, an ANN consists of three types of layers which are classified as input, hidden, and output (Figure 1a). The purpose of the input and output layers are self-explanatory while the actual information is processed in one or more hidden layers. The number and type of hidden layers that is needed is not fixed and is dependent on the particular problem being considered. While the actual mathematics are well established, it is beyond the scope of the present paper to cover this in detail. However, in simplest terms, the inputs are used to construct a vector of input values that is then multiplied by a vector of weights. This weighted sum is then transformed using a nonlinear function. The most common transformation is what is referred to as a rectified linear unit (ReLU) and was also used in our work. The modified data is then sequentially processed in each layer of the hidden nodes (Figure part will be covered in a future article. ANN/MIMO SystemDesign

Introduction There is a great deal of interest regarding Industry 4.0 implementation and the use of Artificial Intelligence/Machine Learning (AI/ML) in the manufacturing community. Given the current economic climate in the forging industry, the combination of machine and human intelligence will be necessary to address technological challenges and to remain globally competitive. While the forging process poses numerous obstacles/challenges to automation generally, AI/ML has demonstrated to provide improved productivity, real time monitoring, and gains in process efficiency, among others. However, while Artificial Neural Network (ANN) technology inside of AI/ML is relatively common in CNC machining, similar forging applications are effectively in the embryonic stage. A few forging-based ANN applications can be found in the literature [1-3], but these efforts have focused on cavity design and developing workability maps rather than predicting and controlling process/part attributes. Moreover, there have not been any reports of efforts to link ANN software and sensors in a holistic “Big Data” system. Consequently, work is needed to advance the use of this technology in the forging industry. One feature of Industry 4.0 is that it requires the synergy between sensing, control, digital technology, and communications on the shop floor. While ANNs have taken center stage, it is likely that the use of machine intelligence technology in forging will also grow as Industry 4.0 is on track to be adopted worldwide. Complex multiple input/multiple output (MIMO) systems are complementary to an AI/ML system. In a sense, AI/ML is software-based whereas MIMO uses a sensor array that enables a linkage between virtual and physical environments. Moreover, in a forging environment, the addition of a sensor array enables other desirable outcomes related to Industry 4.0 and the use of an ANN.The first is to perform automatic collection of input data which would be processed in the ANN. A second reason is to enable the use of self-learning ANN systems which can autonomously improve their performance as new data is collected and entered. A third function is to act as a data

FIA MAGAZINE | NOVEMBER 2022 78