Artificial intelligence can bring significant maintenance savings to SMEs

Undetected failures are the most expensive ones

Manufacturing companies are looking for solutions that automate and reduce maintenance costs. Traditional vibrodiagnostic methods can be too late in many cases. Taking readings in the presence of a diagnostician once in a while may not detect a fault in advance. Mr. Zykmund from the Czech potato chip company knows this. A pump misalignment due to a frame buckling, repeated several times already, has already cost a lot in equipment breakages and subsequent production outages.

Artificial intelligence in maintenance is no longer science fiction

A maintenance manager is looking at the latest news on his phone at home before going to work and receives an alert that there is a sound anomaly on one of his helium compressors. At the same time, the AI library identifies a loose bolt on the compressor’s frame. The manager picks up the phone and calls his colleague, who is about to start his day shift, to make a targeted inspection of the machine parts. With timely intervention, the misalignment in the equipment is detected. This would have caused the compressor shaft to fail and production to be down for hours.

The most common shortage – qualified maintenance workers

Stories from companies that have embarked on the digital journey are no longer just science fiction. They are real examples of how SMEs are coping with the lack of skilled labor on the market. “We used to have one mechanic-maintainer who regularly went round all the machines and diagnosed their condition by listening to them. But such experienced colleagues are retiring and no new ones are coming in,” I often hear from company representatives who come to Neuron Soundware.

The helium compressor example concerns a Czech manufacturer of transmission equipment for the automotive industry. In this case, a failure without early identification would mean three major things: replacing the entire piece of equipment, the spare part of which may not be in stock right now because it is expensive to stock replacement equipment. Next, the devaluation of the current pieces of gear in production, and thus the discarding of the entire production run. Last but not least, it would represent up to 16 hours of production downtime. The losses would run into tens of thousands of euros.

The trend is to monitor machines remotely in real-time

Such a critical scenario is not possible if the maintenance technology is equipped with artificial intelligence in addition to the mechanical knowledge of the machines. It applies this knowledge itself to the current state of the machine and is able to recognize which anomalous behavior is currently occurring on the machine and, based on this, send the corresponding alert with precise maintenance instructions. This is used today, for example, by manufacturers of mechanical equipment such as lifts, escalators, and mobile equipment. At an international airport in Germany, we monitor a moving walkway in this way, detecting damage to the wheels and reporting the need to replace them.

Artificial intelligence helps at various stages of production

But predictive maintenance technologies have much wider applications. Thanks to the learning capabilities of artificial intelligence, they are very versatile. For example, we are able to assist in end-of-line testing. We have carried out several projects in the automotive sector for manufacturers of air conditioning units or fuel pumps. We are now successfully identifying defective or potentially unreliable wheel rims, where cracks invisible to the eye appear randomly. John Harper of Maxion Wheels particularly appreciates the automated diagnosis of the fault and the clarity of the instructions, which tell him exactly what has gone wrong with the product. The second area of application lies in the monitoring of production processes. We can imagine this with the example of a gravel crusher. A conveyor delivers different sized pieces of stone into grinders, which are to yield a given granularity of gravel. Previously, the manufacturer would run the crusher for a pre-determined amount of time, to make sure that even in the presence of the largest pieces of rock, sufficient crushing occurred. Now, with the audio diagnostics application, he can have the artificial intelligence “listen” to the size of the gravel and stop the crushing process at the right point. This means not only saving wear and tear on the crushing equipment but more importantly, saving time and increasing the volume of gravel delivered per shift. This brings great financial benefit to the producer.

The greatest savings are in companies with a high number of identical assets

When implementing predictive maintenance technology, it doesn’t matter how big the company is. The most common decision criterion is the scalability of the deployed solution. In companies with a large number of mechanically similar devices, it is possible to quickly collect samples that represent individual problems and from which the neural network learns. It can then handle any number of machines at once. The more machines, the more opportunities for the neural network to learn and apply detection of unwanted sounds. Semiconductor manufacturers in Malaysia know this. They are now deploying predictive maintenance technology on all of their vacuum pumps that are essential for production, helping them cope with the increasing demand for computing units such as those in cars.

The future of predictive maintenance: accessible and ubiquitous

Due to the cost of computer technology and data processing, condition monitoring technologies are usually designed for larger plants rather than for workshops with a few machine tools. However, as hardware and data transmission and processing get progressively cheaper, the technology is getting there too. So even a home marmalade maker will soon have the confidence that his machines will make enough produce and deliver orders to customers on time and not ruin its reputation.

In the future, predictive maintenance will be a necessity, not only in industry, but also, for example, in larger electronic appliances such as refrigerators and coffee machines, or in cars. For example, we can all recognize a damaged exhaust or an unusual sounding engine. But it’s often too late to drive the car safely home from a holiday, for example, without a visit to the workshop. With the installation of an AI-driven detection device, we will know about the impending breakdown in time and be able to resolve the problem in time, before the engine seizes up and we have to call a towing service. This case, by the way, happened to a friend of mine and was one of the main reasons why we started developing this technology here in the Czech Republic.

The text was written by Pavel Konečný for Trade News 9/2021