This website uses cookies primarily for visitor analytics. Certain pages will ask you to fill in contact details to receive additional information. On these pages you have the option of having the site log your details for future visits. Indicating you want the site to remember your details will place a cookie on your device. To view our full cookie policy, please click here. You can also view it at any time by going to our Contact Us page.

Close

Making the best of AI

28 June 2024

Suzanne Gill looks at the growing role that artificial intelligence (AI) has to play in ensuring successful maintenance strategies and finds out how to make the best use of the technology.

One of the biggest challenges facing food and beverage manufacturers today, from a maintenance perspective, is replicating human intuition. “A skilled operator working in conjunction with an effective preventative maintenance strategy will be able to identify an issue with an oven or conveyor or know when to change a specific tool and take appropriate action before the potential fault seriously impacts production,” said Patrick Dion-Fehily, Digitalisation Specialist at Mitsubishi Electric’s Automation Systems Division. “However, the food and beverage industry is currently facing a skills shortage, from which maintenance engineers are not immune. AI technology has a role to play here, helping alleviate some of the burden facing maintenance teams by passing on the knowhow of skilled workers onto the wider team.” 

So, what is Patrick’s advice for food and beverage production sites looking at adopting AI as part of their maintenance strategies? “You need to have a crystal-clear understanding of what you wish to achieve. Without this, it will be very difficult to identify where AI technology can be implemented in order to attain the optimal result.”

Patrick went on to point out that one of the most time consuming but crucial factors when installing AI into a system is to have the appropriate data set associated with the application it would be used for. Without this, an accurate AI model would be difficult to create and would not, therefore, output the desired result. 

As a result, the first thing a food or beverage manufacturer should do is work with their AI software provider to assess what data is available on the factory floor and whether it is fit for purpose. “Once this has been achieved, you can create the data set, and then create a task based on how you want the data to be distilled, analysed and visualised, before it is fed back into to the control system to start delivering the desired output,” concluded Patrick.

Predictive maintenance
AI is growing in importance in Predictive Maintenance (PdM) strategies, where its machine learning subset can be used to significantly reduce manual effort in processing data. However, according to Chris James, Business & Portfolio Management, Condition Monitoring Systems at SKF, AI is not necessarily the most critical element of PdM. He said: “There is a lot of digitalisation hype around PdM, but the basic principle of the strategy is to perform maintenance when it is predicted to be needed.”

The PdM process involves monitoring condition indicators (CIs) that reflect the physical health of assets, detecting anomalies, predicting a failure, generating a recommended action, and then implementing that action in the form of a maintenance task. It is the last step that limits the level of automation (or AI) that can be applied to PdM, until the day comes when a Star-Wars droid can replace a coupling or tighten a mounting bolt.

A more traditional strategy is preventive maintenance, which has the disadvantage, and cost, of taking apart perfectly healthy machines on a regular basis. However, such solutions will always be required in highly regulated industries. “Today, it is still unimaginable that maintenance would be performed on a passenger aircraft only when some AI algorithm says so! Similarly, in the food and beverage sector, there are some tasks that will always be mandated by food-safety,” explained Chris. There is also reactive maintenance to consider. For some assets, simply letting them fail will still make sense.

Chris’s advice to food industry engineers is to carefully filter the sales pitches about AI and PdM being able to eliminate all your maintenance woes, and instead objectively assess what is the best balance of all three possible maintenance strategies. “Predictive maintenance, and some form of AI, will certainly have a place, but make sure it is being applied to the right assets, at the right level,” he said.

The most important step of a PdM project will be to choose the right CIs in the first place. “The old phrase ‘garbage in equals garbage out’ is still as relevant today as it was with the first computers,” argued Chris. When the CIs have been identified it is then possible to decide on the best way to leverage AI assistance in an operation.

Today, there are two major tactics used to execute a PdM strategy – condition monitoring (CM) and data analytics (DA). 

CM on rotating equipment uses vibration as a CI and involves specialised sensors and systems. Crucially, it also uses signatures to ‘look inside’ the machine to identify faults at an early stage, long before the whole asset is shaking itself to pieces. Temperature is used as a last resort CI – by the time an asset is too hot, it is probably too late.

DA takes a different approach and uses existing sensors in the plant automation system, providing many different scalar values such as temperature, pressure or flowrate. The data resides in enormous databases within a plant historian. Software algorithms are used to analyse the many ‘time-series’ trends, which can reveal anomalies that indicate that maintenance may be needed. AI, or more specifically machine learning (ML), is used in both approaches to automate the tedious task of anomaly detection. “In CM, physics-based rules are then applied to diagnose a potential problem, requiring little context, and a relatively small data-set can determine a specific issue and corrective action,” continued Chris. For example ‘Asset 1234, Motor ABC, Bearing XYZ – outer race defect symptoms – change lubrication interval.’

With DA, ML uses statistical models to detect anomalies. These work best when there is a lot of (good) data. The data is scanned for patterns and anomalies, and any diagnostic rules are case-based, which requires context. The approach is more easily scalable than CM, but each result is less specific. For example ‘Asset 1234, Motor ABC – unusual current draw – inspect motor.’

“The best solution is once again to have a good balance,” said Chris. “Simpler machines may only need anomaly detection, while more important assets may also benefit from a diagnosis. So, you need to pick the right CI, and then cut the cloth to suit your machines.”

A force multiplier
According to Richard Jeffers, Service Solutions & Technical Director at RS, AI can be a force multiplier to an organisation that is already well advanced on the maintenance maturity journey. “If you have a robust condition monitoring programme in place, and the right ways of working to respond to leading indicators of failure, then AI can help you deliver this at scale over multiple assets,” he said. “The use of generative AI can also help identify trends in engineers comments on computerised maintenance management system (CMMS) platforms in a way that is impossible to scale with human analysis only.” Richard’s advice is to focus on getting the fundamentals of PdM in place through effective work management and spare part management and a solid approach to condition monitoring on critical assets. Then look at AI enabled PdM decision support tools.
 
“Without the basics in place, AI decision support tools will be an expensive waste of time,” he said. “You have to have the basics of Cleaning, Inspection, Lubrication & Tightening (CILT) in place; robust criticality to ensure you are focusing on the right assets; strong work management processes so you can respond at pace to alerts; good management of engineering spares to reduce Mean-Time-to-Repair (MTTR) and the right leading indicators of failure being monitoring through your CM programme. Once these are in place, you can deploy AI assisted systems to expand best practice at scale.”

Boosting uptime
Keith Thornhill, head of food and beverage at Siemens Digital Industries, argued that while condition monitoring is not a new concept – with each piece of equipment on a production line representing a potential goldmine of data, ranging from performance output and power consumption to operating temperature – AI is creating new possibilities for how it can be used to boost uptime.  
“Rather than relying on manual checks and visual cues, cloud-based dashboards can give users an overview of the condition of every piece of equipment in a plant. And by augmenting with AI-powered platforms, it is possible to analyse the data and predict exactly when performance might drop or issues might arise with a level of accuracy previously unachievable,” he said. 

Keith’s advice for engineers in the food and beverage sector is to start small before expanding the use of AI across operations. Manufacturing processes in the sector care often highly complex and involve multiple stages, so pick one part of the process, demonstrate good return on investment and build out your digital transformation from there.
“AI is an exciting, transformative opportunity. But it’s vital to treat it as one of the tools needed to achieve individual objectives, rather than a solution for all outcomes,” he said. “AI-assisted maintenance solutions can only be truly effective if underpinned by an operational framework that allows them to flourish. 

“Although predictive maintenance can provide a clear picture of what maintenance is needed and when, human expertise remains essential in acting on these suggestions and delivering the necessary adjustments. The technology also needs constant performance monitoring to operate effectively. Technicians should complete regular assessments – including algorithm refinement and stress testing – identify any weaknesses and stamp out inconsistencies that could lead to incorrect data analysis.”

A recipe for success?
“The success of predictive maintenance strategies derives from a combination of technological tools, as well as organisational and mindset changes,” argued Marielle In de Braek, Global Business Development Manager, LifeTime Solutions Food & Beverage at Schaeffler. From a technical angle, using advanced AI tools is fundamental to this success as AI-based predictive maintenance has the potential to reduce downtime, improve production efficiency, extend the lifespan of equipment, maintain the quality and consistency of production. 

Marielle pointed out that one of the strengths of AI lies in its ability to efficiently process and analyse huge volumes of data. “For machines used in food processing, countless data points relaying information about vibration, temperature, pressure, operations and processes are produced each day. Engineers need to adopt solutions that are able to gather, store and process data in order to enable quick value creation. AI can accurately monitor and interpret this data, identifying anomalies, patterns or predictive factors that only a very experienced human technician would see,” she said. “And it is also worth mentioning that the shortage of qualified and experienced technical personnel able to inspect and assess machinery health might become critical in the near future. Investment in products that enable AI-supported maintenance are also relevant in this respect.”

However, Marielle point out that technology will only contribute marginally if there is not also a clear change in mindsets to embrace these new tools and harmonising them into daily work and processes. “Engineers considering adopting AI for predictive maintenance should consider a holistic approach, assessing how to transition from traditional workflows and processes to an AI augmented daily job,” she advised.

Building an in-house solution is a very complex task that requires a mature IT department experienced in securely executing the development, operation and maintenance of highly sophisticated software, as well as data pipelines.

“The recent revolution in the AI Large Language model and the availability of versatile basic models that are trained on huge amounts of data and can be applied to many new downstream tasks, are now generally available,” pointed out Marielle. “Ad-hoc solutions for simplifying some reporting-related tasks from maintenance operators will also be on the market soon.” 

In conclusion, Marielle pointed out that two main aspects are fundamental to AI adoption, from a technological standpoint. The AI journey begins with ensuring access to the right data, by acquiring detailed, accurate and diverse data that the AI system can learn from. So an important first step must be to invest in quality sensors and data acquisition systems to help capture the best data upon which your AI systems are able to feed.

Conclusion
Remember that AI is not just a technological shift, but is also a cultural one that requires a change in mindset right across an organisation. Creating an organisational culture that supports this change is an important key to ensure success.


Contact Details and Archive...

Print this page | E-mail this page