Predictive maintenance

 Confessions of a process engineer.

Moving pumping station maintenance to digitalisation

With lean, identify wasted steps these are parts of the process that are not valued by the end customer. For example with sewage pumping stations a maintenance technician might drive out to the pumping station to inspect the pumping station and find that no repairs are needed. No maintenance is required. They might drive out two or three times. One of those occasions maintenance intervention is needed and the work completed. The end customer only values that the pumping station continues to work and has never broken down. To them 2 out of 3 of those visits are wasted steps. Replacing physical site physics physics with remote monitoring and an algorithm that recognises when a maintenance intervention is necessary and raises a job for the technicians to drive out Remove the wasted steps. And that is lean.

He’s nitty-gritty. Collected data from a wide range of sources. I collected the telemetry data is remotely monitoring the asset. I collected customer complaint. Data has come through to the call centre. These are two very different types of data. The first is Merickel or can be easily converted into miracle for example start and stop Can be converted to one and zero And so can easily be put into a spreadsheet and then a data program. The second is the log of a phone call between a centre agent and a customer. So this is qualitative it describes the condition from the perspective of the customers experience. I also spoke to mechanical and electrical technicians to gain his best intelligence on the assets from them. And I drove out to visit pumping stations and open them up to read the logbook inside them where technicians that had visited the sites had recorded their findings. This last piece of data Qualitative data and quantitative data that aligns with the first piece that telemetry data. They both have dates.

Customer data has dates. By putting the data together mapping condition and performance result For example a Pamper has stopped On a specific date and after a certain amount of time, a customer has called a complaint about flooding. And a visit immediately prior to the stop or after the stop, but before the flooding would give me descriptive data as to cause prior or the current condition after the stop before the flood. With this and intelligence from mechanical and electrical teams I could create a model Whereby the telemetry data had a pattern of telemetry data is informed by the qualitative data that you match up with.

The testing phase Historical incident whereby I would predict from the quantitative and quantitative data performance behaviour failure would occur. And then I was able to check if it had occurred or not with incredible accuracy. Been very successful with predicting past events it was safe to move to applying methodology to a small number of pumping stations.

At this stage, I’m still doing the whole process manually and it’s quite painstaking and time-consuming. My boss called a cottage industry. But once the proof of concept was proven with a small number of pumping stations Engaged a data scientist I was to hand him a brief and the pattern of quantitative data and what type of maintenance intervention was required for this date of profile. So if the data was indicating that a pump failed, then the maintenance intervention would be to order a new installer new pump. If the data profile indicated that an NRV had failed The maintenance intervention Would be To inspect the NRV unblock it necessary if it is blocked or to border and replace   A new NRV.

They say that genius is 99% perspiration on bond inspiration and I find that is true.

The next stage was to run both my method, the cottage industry and the algorithm written by the day the scientist to reproduced what I was doing, but we have a computer program, Simultaneously. We run both methods and that the end result was the same for each method.

Once that was proven and reliable, the system was rolled out to 5000 pumping stations.

And that is lean.

Lean not only uses the resources required for site visits it reduces the carbon associated with acid maintenance as there is less travel involved and also then targets maintenance more accurately the times that has been used up previously visiting a site to find it. It did not need a maintenance intervention can be used to visit sites that do need a maintenance invention. There’s less chance of being busy with a site less in need of repair well another site is causing customer dissatisfaction.