Last week, we presented the “bathtub” function which shows the failure rate of assets over their lifetime. When this function is approximately known for an asset, a new service model can be introduced to avoid long downtime, dangerous operation modes and untimeliness of failure. The first model that exploits this knowledge is preventive maintenance.

As the bathtub curve showed, a group of identical assets, under the same conditions, does not experience failure at the same time. The failure rate is low for a long period, after which it starts to increase towards the end of the lifetime. As the name says, preventive maintenance replaces your asset precautionary, generally so that 99.7% of all failures occur after replacement. Generally these replacements are routinely performed based upon uptime or usage. For example, most bearings are replaced after 20.000 operating hours regardless of the speed, while your car’s timing belt needs to be replaced after 80.000 driven kilometers regardless of the age.

In most situations, Preventive Maintenance is beneficial over the Run-to-Failure model. By replacing parts before they break, you practically never experience unexpected downtime. Furthermore, due to the avoidance of catastrophic failures, operating your asset is safer and maintenance becomes simpler. 

However, as replacement is planned and scheduled, Preventive Maintenance requires a costly scheduling infrastructure. Large management programs are necessary which track the schedule for all parts of all assets and ensure enough mechanics are available to execute the maintenance. Furthermore, with Preventive Maintenance you replace parts while the failure rate is low. This can shorten the lifetime significantly, especially when the variance of failures is large. Not only does this cause additional costs in assets, it also results in more maintenance work over time. 

At Amplo we try to diminish these disadvantages. The behavior of your assets changes as the failure rate increases. These changes are reflected in data coming from sensors that measure various states of your asset, such as temperature, speed, power, pressure, etc. Our state of the art machine learning algorithms identify how your asset’s behavior changes towards failure. Thereby we acquire the failure probability and an estimated time of failure. With this information we can optimize your asset lifetime, resulting in less frequent maintenance and correspondingly less maintenance costs. This will further decrease the pressure on your complex management systems. 


And by the way, pretty soon we’ll write an in-depth article of how these methods work 😉