AI & Data analysis for predictive maintenance of sewerage pumping system

The International Asset Performance 4.0 Conference will be held from 15 to 17 September 2020 in Belgium. On this occasion, several companies have come together to organize a Hackathon. Among them is Aquafin, a Flemish specialist in wastewater treatment. Their objective is to improve the daily maintenance management of their pumping stations, using data combined with artificial intelligence. The company thus proposes to the participants to find the best possible algorithm to predict the risks of failure of their pumps, and/or to provide a data visualization system, all this based on data from four pumping stations over the whole year 2019, including a history of maintenance interventions, and pump water flow and hydraulic motor intensity data. This challenge is an opportunity for us to confront the CARL|Berger-Levrault research work on AI for equipment management with data from real life. We pre-process them, then constitute operating episodes in the form of time windows.

Pump station behavior clustering

One of the main difficulties with AI is to visualize data in the right way. We projected the data in multi-dimensional spaces, to find correlations and identify representative sets of normal or abnormal states. We are actually exploring the use of co-occurrence matrix to discover correlations and extract features from raw data.

As shown in this video, we animate the data to understand its dynamics and thus observe, model, and understand the transition from a normal state to an abnormal state: this is how we can anticipate and predict failures. This method will allow Aquafin technicians who work on the pumps to visualize the state of health of the pumps so that they could act accordingly if the data ever showed abnormal behavior.

This work is not specific to pumping stations: it is adaptable to a large number of equipment, regardless of the field. Numerous improvements are also under consideration, in particular to automate as much as possible the process of creating these correlations, as well as the detection of failures. 

More ...

Retour en haut