According to the National Association of Manufacturers, for every dollar spent in manufacturing, approximately $3 is added to the economy. The economic influence of the manufacturing industry makes the case for the need to continuously optimize industrial processes to achieve growth and the industry has risen to the challenge.
Statistics show that the manufacturing industry has become one of the largest adopters of digital transformation solutions to drive growth. The adoption of digital transformation tools such as discrete event simulation modeling has played important roles in improving manufacturing outcomes through accurate data analysis, risk-based scheduling, and planning. But Industry 4.0 promises more in terms of automating decision-making and reducing the work-load of technicians and analysts across the industry –here, artificial neural networks have crucial roles to play.
Improving the Performance of Machine Systems
Understanding machine utilization processes have become the foundation for developing predictive maintenance strategies, and defining machine performance. With this knowledge, manufacturers can develop systems to optimize future machine performances to increase throughput and eliminate downtime.
ANN can expand the capabilities of simulation models to predict the mechanical properties of the equipment used within the shop floor. One example is the prediction of the mechanical properties of a friction stir welding (FSW) process. The FSW process involves the use of a rotating non-consumable tool that plunges into a work-piece to produce the heat needed to weld a joint. Data sets or parameters that define the FSW process are the tool travel speed, the tool’s rotational speed, and the grain size and dislocation density of the welded material.
Optimizing the FSW process and the quality of its welds requires extensive control of the mechanical properties associated with using an FSW machine. Thus, an understanding of the relationship between the different variables associated with the FSW process and the weld quality it produces is required. ANN provides a means to predict the mechanical behavior of FSW joints and processes once the required data or parameters are fed into the neural network. In this example, a simulation model of the FSW was developed using a neural network. The historical data used to train the neural network included tool rotation, travel speed, tensile strength values, welding direction etc.
The neural network was able to predict with significant accuracy, the mechanical properties of the FSW process, and the strength of the welded joints under specific operational conditions. Hence, manufacturers can determine the outcomes of FSW cycles to develop new operational techniques to improve productivity.
Improving Facility-wide Decision Making Processes
The application of ANN within the manufacturing industry is not limited to optimizing machine performances. A major aspect of its application is improving the decision-making processes associated with facility-wide operations. Examples of such operations include selecting the optimal choice between production lines with multiple workstations or evaluating the cost and time-based effects of diverse production variables to a production cycle.
For example, a manufacturer interested in replacing conventional material handling systems with automated infrastructure must determine the best option for a facility’s peculiar requirements. The factors that influence the functionality of the material handling process include the status of the delivery workstations, factory layout, the speed of alternative automated material handling systems, as well as, the navigation capacity of the proposed options.
Although a conventional discrete event simulation model can help, knowing how to consider all the above factors and construct the rule to select the best automated system is a difficult, time consuming process. Integrating ANN into simulation models replaces the need to construct the rules that capture the complexity of manufacturing operations. The neural network is then continuously trained using simulated data to improve its decision-making accuracy levels. Utilizing an ANN ensures that the process of building complex rule-based logic is replaced with a more accurate system that optimizes manufacturing operations and decision making.
Conclusion
ANN simplifies the decision making process at the machine, process, and facility level within the manufacturing floor. Neural networks increase the accuracy of a simulation and digital twin model’s evaluations and analytical capabilities. The ability to constantly train neural networks using synthetic training data ensure that ANN algorithms are capable of adapting to diverse situations just like the human brain.