Machine Learning in Industry

This introductory chapter describes the techniques of machine learning. Primarily the concept of machine learning in the context of artificial intelligence and data analytics is explained. The application process of the above to big data is introduced. Classification of machine learning approaches is described. Then some of the variedly used statistical and artificial intelligence-based machine learning techniques are described in brief. The techniques discussed include decision tree, linear regression, least square method, artificial neural network, clustering techniques. The concepts of deep learning are also introduced.

In this study, a systematic comparative study is made between selected ways for identification of a neural network model for the prediction of the residual stress of steel based on the non-destructive Barkhausen noise measurement. The compared approaches are the deterministic forward selection with and without filter and a stochastic genetic algorithm-based approach. All the algorithms make use of the extreme learning machine as a model basis. The main objective is to propose a systematic procedure for identifying a prediction model for the considered system. The results of this study show that the algorithmic approach might be considered necessary not only to reduce the effort for model selection but also to select models with high prediction performance. It was also found that the genetic algorithm proposed earlier by the authors is applicable for selecting a well-generalizing model to the system, but the performance of the deterministic selection techniques is also comparable to the genetic algorithm.

Optimization techniques are applied widely in iron and steel making process to solve complicated process-related problems. These methods and the models created through them are regularly used in this field to find out the optimum operating conditions in terms of cost, quality, quantity and effectiveness of the process. The various blast furnace parameters like burden distribution, oxygen enrichment, productivity improvement, composition of the top gas, quality of hot metal production, etc., are very difficult to effectively optimize. In recent times data-driven models of diverse nature have been quite successfully applied for this purpose, where evolutionary approaches have made a significant impact in simultaneous optimization of multiple number of objectives in problems related to the iron and steel making industry. In this chapter implementation of various evolutionary strategies are discussed, which are recently applied in this domain to tackle some real-world problems.

Machine learning has come a long way since its inception. It has totally revolutionized the industrial scenario. For decades, there is clear evidence of the usage of sophisticated digital control and monitoring systems by industrial operators. These entail a multitude of sensors with defined functionalities. Through the adoption of machine learning, their handling becomes a lot easier. In this chapter, the different strategies adopted through machine learning are being highlighted. Additionally, the challenges of effective integration of industrial data, including that from sensors, for standard ML are outlined. Apart from this, this chapter appraises readers about predictive maintenance; accompanied by recommendations.

The actual cause of sliver defect is difficult to determine, since the defect usually reveals itself after the rolling process (hot/cold) is complete. The genesis of sliver defect in cold rolled steel sheets is investigated in this work using two popular computational intelligence tools used in data mining, namely, rough set and fuzzy set theories. A substantial amount of data starting from the steelmaking stage to finish rolling of the product has been collected with the aim of extracting useful knowledge about plausible cause(s) of sliver formation. While rough set theory helps to select the important variables to which the cause of the defect can be attributed in the form of rules, these rules are given a linguistic form through fuzzy membership functions. A rule base thus evolves in the form of a fuzzy inference system constituting a few important variables, which serves as a perceptive model for predicting the severity of sliver defects in cold rolled steel. Validation of the fuzzy system is done using actual industrial trials.

Materials science research begins in laboratories with testing the properties of metals and their alloys, the properties of the material depending on the type of additives and microstructure, as well as the changes in these properties taking place under the influence of processing. The next step is modeling and simulation of processes to investigate the possibility of their control and monitoring under production conditions. Some studies relate to an ongoing process, and then the research focuses on quality control of the process, optimization, and detection of irregularities and product defects. At all stages of research, it is possible to apply the methods of machine learning to the extent chosen by the analyst or expert. These methods can be used to obtain knowledge about occurring phenomena, research planning, and designing of production processes (in accordance with the 4th paradigm of science), but they can also be data-driven models given the possibility of autonomous control of a selected aspect of production (in accordance with the idea of the 4th industrial revolution). This paper presents an overview of ML methods based on examples taken from the field of materials science discussed in terms of materials–processes–knowledge formalization.

The Universal Approximation Theorem provides the theoretical basis for perceptron-like architectures to represent the functionality of complicated mathematical functions to any desired degree of accuracy. Among the most complex of such functions are the governing equations of physical processes like the Navier-Stokes and Maxwell’s equations. Accurate representation of complex physical phenomena through numerical simulation of such governing equations is a challenge, and it is a concomitant challenge for Artificial Neural Networks (ANNs) to learn the functionality of these equations from data generated from such simulations. There is an intense practical value of such analysis—most sub-processes in the industry are describable by approximate versions of these equations, their solution in real time will enable significantly more informative and precise monitoring, control and prognostics of running processes. Arrays of sensors installed at the physical boundaries of process domains can provide the crucial inputs to ANN-like architectures that can transform these isolated values into detailed field conditions. Provided these ANN-like mechanisms can exhibit the following properties—that they respond in process real time, their solutions are nearly as accurate as of the offline numerical simulations of the governing equations from which they learn the functional relationships, and importantly, they can map a few score inputs into around two orders of magnitude more number of outputs. The development of such architectures will open entirely new vistas of application of ANNs to modern industry. Here, we present convolutional-NN-like architectures based primarily on transpose convolutions and other design features that satisfy all the three crucial properties. These are demonstrated on two different application domains of the reduced Navier–Stokes equations, containing high nonlinearities and abrupt discontinuities including shocks.

Deep learning has influenced almost all major domains of science, technology and engineering fields. The deep learning revolution started with the ground-breaking accuracy obtained in a computer vision problem. Machine vision-based inspection has been one of the pioneering applications of computer vision for industrial applications. The adoption of deep learning for machine vision applications took some time, and the current adoption rate though is satisfactory, it is observed that there is still a long way to go. The contents of the chapter is intended for beginners and managers who are evaluating application of deep learning techniques for vision-based automated inspection. This chapter presents detailed insights of merits and limitations of deep learning techniques for automated inspection tasks, especially in comparison to non-deep learning route. The various aspects of commissioning such as important pitfalls to be cautious about before choosing deep learning, deep learning software, deep learning hardware, types of deep learning networks and their inferences and potential applications in various types of industries are also discussed.

State-of-the-art infrastructure, excellent computational facilities and ubiquitous connectivity across the industries have led to the generation of large amounts of heterogeneous process data. At the same time, the applicability of machine learning and artificial intelligence is witnessing a significant rise in academics and engineering, leading to the development of a large number of resources and tools. However, the number of research works and applications aimed at implementing data sciences to problems in process industries is far less. The proposed work aims to fill the niche by proposing Artificial Neural Network (ANN)-based surrogate construction using extremely nonlinear, static, high dimensional (32 features) noisy data sampled irregularly from inlet and outlet streams of hot rolling process in iron and steel making industry. Though ANNs are used extensively for modelling nonlinear data, literature survey has shown that their modelling is governed by heuristics thus making them inefficient for use in process industries. This aspect is of high relevance in contemporary times as hyper-parameter optimization, automated machine learning and neural architecture search (NAS) constitute a major share of current research in data sciences. We propose a novel multi-objective evolutionary NAS algorithm to optimally design multi-layered feed-forward ANNs by balancing the aspects of parsimony and accuracy. The integer nonlinear programming problem of ANN design is solved using binary coded Non-Dominated Sorting Genetic Algorithm (NSGA-II). ANNs designed for the hot rolling process are found to demonstrate an accuracy of 0.98 (averaged on three outputs) measured in terms of correlation coefficient R² on the test set. The successful construction of accurate and optimal ANNs provides a first-of-its-kind model for the hot rolling process in the iron and steel making industry. The proposed method can minimize the chances of over-fitting in ANNs and provides a generic method applicable to any kind of data/model from process industries.