How Making Your Machine Learn Patterns and Infer from Data can Change the Manufacturing Shop Floor
We’ve seen numerous real-world applications of Artificial Intelligence (AI) making a positive impact in our everyday lives. This can range from AI-infused items such as autonomous vehicles and smart home devices that cleverly anticipate our personal needs and take self-control, to services such as online purchasing platforms that capture our transactional patterns, predict and automatically pushes recommendations of just what we’ll be interested in based on algorithms. The main enabler of this intelligence is Machine Learning - an application that uses patterns and inferences to teach systems to perform specific tasks without being explicitly programmed.
On the food and beverage manufacturing front, we can expect to see several benefits of AI applications as well. There are several areas of AI and we attempt to classify them in this context:
- Predictive – early detection and warning of issues, inefficiencies and errors
- Performance – first principles analysis and machine learning working together to optimize processes
- Prescriptive – root-cause analysis, optimized solutions and risk-based decision support
- Prognostic – forecast to predict future events, manage risk and optimize cost impacts
Artificial Intelligence in Asset Maintenance Optimization
In a McKinsey study, AI-enhanced predictive maintenance of industrial equipment can generate 10% reduction in annual maintenance costs, up to 20% reduction in downtime and 25% reduction in inspection costs. Rightfully so. Predictive Maintenance leverages on supervised learning, one method of Machine Learning, where data gathered are linked up and goes into “labeled answers”. These inputs set up scenarios where a “correct answer is provided” and tell the machines what to predict.
Equipped with predictions of impending failures, it is no longer necessary to perform inspections based on time schedule. Instead, we maximize the lifespan of equipment parts and replace them as and when necessary – in this case, just right before impending problem occurs. With tight capital and operating budgets, manufacturers are looking to “sweat” existing assets, this predictive maintenance approach translates into significant savings in inspection costs while keeping unplanned downtime to minimum.
Artificial Intelligence in Process Optimization
Manufacturers are often constrained to run a multivariate process on a conservative approach, rather than push near operating limits. Reinforcement learning, another method of Machine Learning, comes in handy especially in such complex operating environment where changing a particular variable may bring about consequential effects on other variables. It works by constantly testing out the dynamic relationships between data and suggesting the most optimum path to take to maximize throughput.
Another example of Machine Learning application is Advanced Process Control (APC), a model predictive control method. APC provides tighter control of key process variables, and effectively decouples interactions that would occur if the same loops were controlled independently by single loop controllers. The ability to perform a series of process response tests and simulations without disruption to the normal process operations, not only means a more economical outcome, it also facilitates agile modification of the process.
Training Makes Perfect
Even in a batch process manufacturing environment, coupling in Machine Learning into real-time production monitoring can be beneficial. With several equipment running at different rates within a line, it can be exasperating when starving or jamming occurs at any point, causing disruption to the entire line.
Tagging each production run at different equipment point in a variety of production rates “train” the system to recognize what a golden batch would look like, and to make inferences based on the actual production rate to identify when and where exactly a potential starving or jamming might occur next. This enables operators with enough lead time to take immediate actions to rectify critical situations.
And it doesn’t stop here. We’ll see more and more practical applications of AI in the food and beverage manufacturing industry. Analytics and AI-driven processes, supported by Machine Learning, will revolutionize the manufacturing shop floor.
Furthermore, at the AVEVA World Summit in Singapore, we'll be discussing these topics with industry peers in the manufacturing space. I encourage you to join me at the event from 16 – 18 September 2019.
Keith Chambers is responsible for strategic direction, commercialization and development for AVEVA's operations management portfolio globally. Keith has over 20 years’ experience in the automation, software and MES business with a focus on manufacturing operations software in the food and beverage, CPG and life sciences industries.