January_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J A N U A R Y 2 0 1 8 2 5 up to 60%enhancement in the first pass yield [7] , a key operational metric. In this case, the analytics approach included a first principle based model and neu- ral network model. This approach pro- vided actionable recommendations in terms of clustering input coils, deter- mining chemistry specification rang- es for input quality, and new process set-points for different coil grades and dimensions. Artificial intelligence (deep learn- ing and machine learning). Deep learn- ing—or neural network—techniques have been used in materials engineer- ing for over a decade in a wide variety of industrial operations, including heat treating, welding, rolling mill, steelmak- ing, and casting operations. In most cases, these techniques are used to build correlations between input mate- rial and process data with quality data. The power of a neural network in accu- rately capturing the thermal and phase transformation kinetics was demon- strated for a batch annealing opera- tion [8] . Subsequently, the effectiveness of this approach over a phenomeno- logical model was demonstrated by optimizing coil dimensions for maxi- mizing operational productivity. The neural network can also be integrated with evolutionary techniques to opti- mize multi-segment thermal cycles [2] or charge dimensions [8] . The reality of noise in terms of measurement errors and lack of avail- ability of upstream process data makes deep learning of industrial data very challenging. However, the prediction error during modeling of industrial data can be effectively reduced by choosing the right algorithms, optimiz- ing the neuron and hidden layers, and superimposing advanced noise reduc- tion methods such as weight decay or a stochastic addition of noise over indus- trial data. In this case, a 0.5% addition of stochastic noise over the industri- al batch annealing data improved pre- diction of yield strength from 89% to 95% [9] . Machine learning as a technology trend is receiving significant attention, where patterns are expected to be ex- tracted from structured datasets. The extracted patterns are leveraged for decision-making, which continuously improves with enhanced usage. These methods are powerful especially in complex industrial situations with multiple variables and no clear under- standing of first principles and the cor- relation between input and output vari- ables. In the context of materials engi- neering, potential applications include selection of recipes, material grades, coatings, or manufacturing processes for a new component based on a histor- ical database of similar decisions. When component portfolios are large, another approach is to enable machine learning algorithms to discov- er patterns from an existing compo- nent database. Similarly, patterns can be discovered for the recipe selections for new parts based on historical expe- rience. A typical solution could entail advanced clustering or a classification algorithm with an architecture to fine- tune the model parameters (Fig 3). It is equally important to under- stand that data-driven intelligence can- not completely substitute for human intelligence, as data-driven intelligence is based on the dataset derived from the current operating regime. If the op- timum operating conditions and right decisions are not within the current datasets, the artificial intelligence will not be able to steer decisions toward the correct operating regime. Blockchain technology. Blockchain technology (such as Bitcoin currency) is a relatively new idea, where an object’s information can be tracked and traced by different user groups at different stages in a secure manner. The next user in the value chain can see the prior informationwithout access tomodifica- tion. Application in the manufacturing world is not yet widely demonstrated. Potential applications include tracking a part or process, supplier compliance, or timestamping critical operations. It is expected that new applications of this technology in the materials and manu- facturing industries will evolve over the next few years. Digital thread/digital twin; Inter- net of Things (IoT) or Industry 4.0. The digital thread of an organization refers to the seamless integration of all the product engineering, manufacturing process, supply chain, quality, human resource, finance, and customer data. Having this access and integration pro- vides an unprecedented opportunity for optimizing the organization at the systems level. This would also enable the organization to run analytics across the production value chain and unravel significant value in terms of productiv- ity enhancement and product quality. Digital twins are the virtual identity of a physical component or system, where parallel to a physical component, vir- tual information is created—including product engineering, supplier, manu- facturing, and quality data. This wealth of information helps organizations run analytics on components and products as well as better manage quality and warranty issues. Industry 4.0—or the Internet of Things (IoT)—is an emerging technol- ogy trend in the manufacturing arena, where convergence of the virtual and Fig. 3 — Broad architecture of a machine learning application in the context of materials engi- neering.