SEPTEMBER 202319By Noemi Guerrer, Senior Manufacturing Engineer and Thad Saunders, Director of Manufacturing Engineering, Clark PacificManufacturing Intelligence (MI) is waterman concept that is till new and in relative infancy. MI centers around the premise of utilizing software in conjunction with existing manufacturing operations to generate deep analytics that was built from the Industrial Internet of Things (IIoT) to drive production. Ultimately, MI is the idea of taking IIoT, digital twins, cloud management and technology, such as Virtual and Augmented Reality (VR/AR) and applying them to the manufacturing floor for improved data and function. These systems form the blueprint to becoming a plant with integrated MI. While simple as a general procedure, implementing MI is not easy and can be costly as it involves deep financial or time investments. The degree ofDifficulty will be determined by a plant's digital maturity. Where to begin then? At the basis of MI is data. Begin by obtaining consistent data from a core product or process.There is a great opportunity in data analytics through Machine Learning (ML) algorithms. These algorithms serve to find patterns in data that human intelligence could not reach or would require deep study. What should the acquired data look like? If the data is coming from a manufacturing facility, that data will often fall under supervised learning in ML which has two categories: Classification and Regression.A classification problem would entail looking at a vast set of inputs that yield a class output. Take a flower for instance. Features such as size, shapes and colors could be noted to classify them as roses, daisies or lilies. Or, categories 0, 1 and 2 respectively. This could lead into asorting process. Will a produced part meet certain quality thresholds? What quality class will be assigned to it? The machine with the ML algorithm would be able to make that decision and accept or reject parts on its own.Suppose the goal is to define if a formula will produce a product with certain performance characteristics like strength or elasticity. Then this is a regression problem. In regression, theoutput may be a gradient value of characteristics such as compressive force for concrete or elasticity for polymers. When prior data is given to a regression algorithm, a model is created tolead to those predictions. Whether it is as simple INSIGHTSMANUFACTURING AI IN DATA ANALYTICS AND MACHINE LEARNINGNoemi GuerrerThad Saunders
< Page 9 | Page 11 >