Predictive maintenance leads to more robust business model
When Alten proposed to a manufacturer of poultry processing equipment to introduce predictive maintenance into their product range, the software service provider initially met with skepticism. “Predictive maintenance sounds nice”, the customer said, “but why should expensive parts be replaced sooner than necessary? Would predictive maintenance not lead to fixing something that’s not actually broken?”
IN HET KORT
Because of their long-term relationship with this machine builder, Alten convinced them that there was nevertheless an opportunity and got a green light to get to work. The trust has started to pay dividends in the customer’s business model. Predictive maintenance now helps the machine builder save costs by minimising maintenance costs and ensuring customer uptime requirements are met within tighter limits.
Don’t spend a euro if it’s not really necessary
The reluctance of the customer to risk replacing parts too early is easy to understand. Even if they are wearing out, use just a little too much energy or are already over their theoretical lifespan there’s a good chance that some expensive components will continue to run for months without intervention. Therefore in general, the credo applies: avoid spending money on replacing parts if it is not really necessary.
Maintaining a poultry processing line is a careful balancing act. In this business it’s all about margins of cents per product and every minute of production time counts. Consequently service level agreements (SLAs) are very common in this world: suppliers of processing lines guarantee uptime. If the lines shut down, they share in the costs.
In order to guarantee production continuity and minimise downtime, many spare parts are held in stock at the abattoirs. Examples include the knives for cutting chicken fillets. If they become blunt or otherwise are at the end of their lives, the abattoir plans an appropriate time to replace them. Equally, bolts and nuts, drive-belts and other low cost items are always available.
The trade-off becomes more complicated when the price of a spare part runs into thousands of euros. A catastrophic problem – such as the failure of a crucial drive – can bring a processing line to a complete standstill. But keeping significant quantities of expensive spare parts in stock at every customer location is also expensive. “These are complex considerations,” says Richard Bernards, business manager at Alten. “Even if you hold everything in stock, there’s no guarantee critical parts will work if they have been sat on a shelf for a year or two.”
Hours of lost production were previously taken for granted
Until recently, the machine supplier sought difficult balance. Cheap components were kept in stock, more expensive parts such as motors were only ‘flown in’ when the need arose. In such a case, restarting a processing line could take hours, but that was accepted as a necessary evil.
Alten proposed to its client to investigate predicting the failure of more expensive parts. If a complex part of the system threatened to break down, it could be replaced at a time when the processing line was stationary. That story turned out to be harder to sell than expected.
Bernards: “The main argument was, we’re not going to spend money if machines aren’t broken.”
In the client’s experience, Altens’ proposal created a new problem: it would incur costs at an earlier stage. Bernards: “They had a different image of predictive maintenance. They imagined, for example, that they would get a message to change a drive when it had been running for three years, irrespective of the actual operating condition of the drive. However, we saw opportunities to predict the time of replacement quite accurately based on data collection”.
Alten proposed to investigate patterns in sensor information from the processing line and use this to make forecasts about whether a component would be likely to break down in the near future. The software house had built up a good relationship with the machine builder over the past years and therefore, despite reservations, gave Alten the green light investigate whether there might be opportunities.
Chicken processing lines generate an overwhelming amount of data
One of the major advantages turned out to be the overwhelming availability of data generated by existing sensors in the relevant chicken processing lines. For example, hundreds of sensors monitor the quality of drumsticks, wings and fillets by means of temperature measurements. If the temperature is too high, the meat is immediately ejected from the line.
Bernards: “There was sufficient data already available. For example, a processing line also makes decisions and performs actions using sensor data. There is also data available about power consumption and motor RPM. Light detectors see the passing chickens and image sensors provide much-needed information for filleting”.
Numerous sensors turned out to be able to provide more information than was necessary until recently. For example, every microprocessor is equipped with a sensor that protects the electronics against overheating. If the temperature exceeds a specific value, the microprocessor switches off.
This information was used to enrich the already available data and thus increase the quality of the predictions. The abundant availability of very diverse data proved to be a huge advantage as the accuracy of the prediction lies mainly in combining, matching and comparing the data. If a drive starts running faster, that may be a consequence of decreased resistance due degraded transmission. It could also be possible that an internal shaft has broken. “A rev sensor only tells you that the revolutions are increasing,” Bernards says. “To know what’s really going on, you need other data.”
Crux is in the smart matching of data
However, in order to be able to forecast future failures properly, smart matching during data collection is not sufficient. “You can’t say: if incident 1 and incident 2 occur at the same time, you have to stop the machine,” says Bernards. “You also need to look at historical data, recognize patterns of catastrophic situations and use them to assess what action to take in the future”. Another advantage in this case was that the customer already captured large amounts of historical information from their latest generation of processing lines, either centrally on a PC or in the cloud.
This eventually led to a system that combines data, learns itself and consequentially continually improves. From the mass of data accumulated an forecast can be made of the chance of failure of a given component. “We train the model to make a recommendation to replace something quickly,” says Bernhards. “We can’t say that something will fail at five past four this afternoon, but we can say, for example, that there is a 90 percent chance that a specific part will fail within now and two weeks”.
Predictive maintenance helps the customer to avoid SLA fines, ensure uptime and keep maintenance costs low. But there is still plenty of room to decide on priorities. For example, the machine builder can propose to the end customer a fixed fee per month to guarantee a processing line availability of 99%. Higher availability levels are less economic as the machine builder incurs increasing costs and runs the risk of replacing parts too early. Bernards: “Predictive maintenance makes it easier for you to choose the right moment to replace components. But you can still make up your mind between decreasing risk or decreasing cost.”
Ultimately Bernhards expects predictive maintenance to bring benefits to both the machine supplier and its customers. “For example, customers get a discount on their SLA if they opt for a software version with predictive maintenance and a higher uptime.”