Working Smarter

AI and Machine Learning in Manufacturing

A key goal of any successful manufacturing operation is a continual drive toward improving the efficiency of the manufacturing process. Traditionally, this has been accomplished through the adoption of lean production principles, waste reduction using the Six Sigma approach, and similar productivity solutions. These systems have been widely incorporated throughout the manufacturing industry and have significantly improved product quality, production speeds, and perhaps most importantly, the safety of those working in manufacturing plants.

These principles remain a key focus on manufacturing floors around the world, but recent innovations in the field of artificial intelligence and machine learning are currently revolutionizing the way we look at manufacturing efficiency.

To understand the fundamental shift that these technologies will bring about in the world of manufacturing, let’s attempt to demystify how they work. Artificial intelligence and machine learning might sound like far-future concepts from science fiction, bringing to mind images of C-3PO and the Terminator, but the technology is very real and you are interacting with examples of it every single day.

When you check the five-day forecast, or you let your phone autocorrect a mistake in a text message, you’re interacting with an AI. When you use a search engine, look at a social media feed, listen to a music streaming service, or see an ad that seems suspiciously geared toward your interests, there is an artificial intelligence operating behind the scenes. We make use of these technologies dozens of times every day and now they are beginning to take hold in the world of industry.

An artificial intelligence is not a machine that can think for itself; this is a misconception that has been driven into the collective consciousness by movie robots and Skynet. So far, no machine has ever had an original thought, and despite what some alarmists say, this is most likely a worry for the distant future. What an AI can do better than any human is analyze mountains of data, find subtle patterns, and use those patterns to make predictions.

When you see a sheet of heavy dark clouds roll in and think to yourself that a storm is coming, that’s not an original creative thought – that’s your brain using a pattern that you’ve learned to recognize to make a prediction. The AI equivalent of this would be a weather station that uses hundreds of sensors to collect thousands of data points like temperature, atmospheric pressure, wind speed, wind direction, humidity, and much more, all in real time. A computer can analyze decades of historical weather data and find subtle patterns that have predictive power. If it sees thousands of examples where one particular combination of these metrics tends to have the same result, it will “learn” to associate that pattern with that result. This is the power of artificial intelligence: using a computer’s capacity to analyze more data than any human ever could and find patterns more subtle than any human would ever recognize.

In the manufacturing world, companies depend on big machines with thousands of moving parts. Every component shifts and grinds and wears down over time. Eventually they fail and that can generate significant losses for a company as a result. In the distant past, the operators of the equipment would attempt to foresee problems by simply paying close attention to the way it ran. A good technician would be one so in harmony with the machine that they could divine an impending failure by the sounds it made or some subtle vibration that only they recognized as new. Over time, the idea of predictive maintenance matured away from this mysterious man-machine oneness toward a more practical approach. Records were kept related to the failure rates of each component and parts were replaced in advance their expected end-of-life.

This process continued to evolve over the years. Eventually, sensors were installed to monitor for specific failure signatures. If a sensor detected smoke or overheating on a critical component, it could sound an alarm, signal a technician, or shut everything down, depending on the severity. In these cases a human would calibrate thresholds for each sensor based on their best guesses, but this manual approach doesn’t account for the highly dynamic nature of machinery or the always changing context of the manufacturing process.

Today, these processes are evolving. In modern manufacturing facilities around the world, hundreds, sometimes thousands, of high-precision sensors collect data from every step of the manufacturing process. This data is continually fed into a central processing system where it is analyzed and compared to predictive models. Patterns that tend to indicate impending failure are instantly detected and reports and generated to help technicians find and repair problems before they result in downtime. In the most advanced setups, when an unexpected failure does occur these systems can review the data collected prior to the failure and attempt to find the patterns that caused it, thereby learning to detect similar failures in the future.

The condition of every critical component is being constantly evaluated and issues are being detected before they occur, no technician-divination required – but predictive maintenance is only one of the many benefits that AI systems bring to industry.

Predictive maintenance means lower labour costs, reduced waste, and less downtime, but all mechanical components still have a lifespan. AI can help a manufacturing company predict with high precision how much longer any given component in its system will last. In the past, the Remaining Useful Life (RUL) of a component was determined by how long they lasted on average, but an AI system can base these predictions on a much more sophisticated model. Sensors can record and analyze every single move that component makes and model the RUL on that one specific component’s unique behaviour, rather than on an average of all components of that type. More advanced systems can even make micro adjustments to the manufacturing process based on what they have observed to reduce wear and improve RUL. Preventing failure for as long as possible is only part of the solution. When a part finally does fail, predicting it in advance will prevent costly downtime.

AI can also be a big help in managing the supply chain. The continual monitoring, analyzing and reporting enabled by AI systems can help a company reduce waste and finely manage the production flow. Mistakes can be detected automatically, which means waste products can be counted with high accuracy, which also results in high accuracy for inventory management. Because every aspect of the process is synchronized, there is no risk of running short on supply; if some material resource is low the system will detect it and replacement orders can be made automatically.

Beyond efficiencies in the manufacturing process, the quality of the resulting product is also improved. In the past, a tiny maladjustment in a piece of equipment could cause a cascading problem, ruining dozens or even hundreds of products before the operators caught on. Worse yet, if the problem is small enough, it might not be caught before the product is shipped. With AI systems, even the most minute issues can be detected and resolved before any waste is generated.

Possibly the most significant improvement AI can bring to the manufacturing world is in improving the safety of plant workers. By detecting problems before they occur, AI can help manufacturing companies mitigate the many dangers that have always been a major challenge in the industry.

Artificial intelligence and machine learning are changing the face of industry. They’re improving the productivity and the profitability of manufacturing companies, they’re reducing the waste they generate, and they’re helping to keep workers safe. As more companies begin to embrace these innovations, products will improve in quality and the costs might even come down. I, for one, welcome our new robot overlords.

Changing Course

N95 masks. Hand sanitizer. PPE. Respirators. Plexiglass barriers. Social distance signage. Rapid Testing Kits. Contact tracing apps. Vaccine against a previously unknown Coronavirus. In January 2020, who would have thought there was a need for manufacturers to ramp up production of any of these items? Or to deal with the grim reality that more body bags were required? But by March, the need for all these items and more was real, and manufacturers struggled to meet demands amid the growing public health crisis of a global pandemic.

Past Issues

March 2, 2021, 11:23 AM EST