Anomaly detection with many possibilities

In the port of Antwerp, ENGIE Electrabel operates ten cogeneration plants that supply steam and electricity to industrial companies. Accurate monitoring and timely detection of measurement deviations of the incoming and outgoing energy flows of the process are crucial, as these energy measurements determine the ecological and financial performance indicators. Automatic anomaly detection offers an efficient solution for this. Unfortunately, this functionality is often a small part of expensive and complex software packages. ENGIE Laborelec went looking for an alternative and found it in Yanomaly from Yazzoom.

Industrial players with a significant energy and heat demand are looking for solutions for efficient and sustainable energy production. In the combined generation of steam and electricity, up to 85% of fuel energy can be put to good use. This makes combined heat and power (CHP) an important link in the energy transition, in which ENGIE Electrabel wants to play a key role. It has therefore been a partner of the customer for more than twenty years in the design, construction and operation of large and small CHP plants across the country. The generated electricity is injected into the grid, while the steam is taken directly by users: factories, but also hospitals or residential sites. In each case, these are crucial installations which must guarantee a high level of operational reliability. However, this is not the only reason why, a few years ago, automatic anomaly detection was introduced on ten large CHPs. “Experience had taught us that this is an efficient way of detecting important anomalies in the supplied energy volumes and the associated performance indicators early and automatically,” says Koen De Bauw, Chief Technologist at ENGIE Laborelec. “This detection can be complex, because the energy flows in one and the same installation can vary greatly. Steam can be generated by parallel units, which in turn can feed users at different pressure and temperature levels. What’s more, energy demand is constantly fluctuating, which can often be significant. In short, accurate and transparent reporting of the ecological and financial indicators of a CHP requires accurate monitoring of the incoming and outgoing energy flows”.

New solution emerges

Until a year ago this was perfectly possible with the anomaly detection functionality of an asset management software package that made use of the available process parameters, such as flow rates, pressures, temperatures, … “The anoamy detection provides a wealth of information,” says Koen De Bauw. “In the CHPs we monitor an average of around fifty parameters per process. Anomalies can indicate a defect, often at a very early stage. If you interpret these data correctly, you can take targeted and timely action and avoid that a defect degenerates into larger problems. We communicate our findings to the operators so that they, with their thorough knowledge of the installation, can carry out the necessary checks and minimise the impact of downtime. The package used in the past by ENGIE Laborelec was no longer commercially available as a stand-alone solution. The license price of a complete software package, of which in the end only a limited part of the functionality was used, was no longer proportionate to the objective”.

CHP plants are an efficient and sustainable solution for the combined production of electric energy and heat (photo: CHP Signpost 2019 – COGEN Flanders)

AI and anomaly detection

Initially, ENGIE Laborelec thought of developing an application itself. “Soon we realised that this would be a time-consuming process with an important price tag,” explains Koen De Bauw. “We thought it would be better to work with a company that specialises in this area. Yazzoom had caught my attention during a research project. After all, this company was developing algorithms that link artificial intelligence to anomaly detection. A blind test confirmed our suspicion that their solution ‘Yanomaly’ was the right tool for our application, even though the package was not yet ready for a real-time application”. “We immediately saw a great complementarity between us and the domain expertise of ENGIE Laborelec in the field of monitoring”, says Jan Verhasselt, CEO of Yazzoom. “Let there be no misunderstanding: ‘Yanomaly’ was not tailor-made for ENGIE Laborelec. But this organisation – just like some other users – has played a crucial role in refining our concept into a marketable product”.

Real joining of forces

Both parties first worked together to integrate the ‘Yanomaly’ algorithms into the cloud environment of ENGIE Laborelec. The result is that the data is extracted in real-time from the cloud database solution (in AWS), while the predicted values and alarms are visualized in the ‘OneBoard’ platform of ENGIE Laborelec. “We also wanted to make the user interface more suitable for the domain experts,” says Jan Verhasselt. “The experts of ENGIE Laborelec gave an important input from their experience with integrating anomaly detection into condition monitoring. This case was also an interesting challenge for improving the robustness compared to poor data quality during live streaming, for example due to disturbed or incomplete data. ENGIE Laborelec’s experience with industrial monitoring activities and data science made it possible to quickly translate technical questions into clear specifications”.

An example of Yanomaly’s integrated solution in the OneBoard platform for one of the monitored units. Deviations between a real process value (blue) and the value predicted by Yanomaly (green) give rise to an increased anomaly score per parameter (purple). The recurrent occurrence of anomaly scores for one or more parameters gives rise to an increase in a general error indicator (BURST) which generates a warning for the user.

Future-oriented tool

Within ENGIE Laborelec, ‘Yanomaly’ is applied today in exactly the same way as the previously used solution. “The big difference is that the application is a very scalable (‘big data’) solution, with various algorithms with which we can go in all directions,” says Koen De Bauw. “We monitor a specific type of process, but have the flexibility to use the anomaly detection or predictive models for other applications with different data volumes, all within our own data management environment. Simple monitoring without artificial intelligence is also possible. The rollout can be done in a simple way, because ‘Yanomaly’ is an application with an intuitive graphical user interface. So you don’t have to be a data scientist to use this system”. Jan Verhasselt responds: “This case indeed illustrates the great ease of use of the solution. Our people have configured a first CHP in ‘Yanomaly’. This training enabled ENGIE Laborelec to realize the other implementations themselves”. “Thanks to the close cooperation with Yazzoom in the implementation, we have an unambiguous understanding of all the steps in the entire data processing process, from the inflow of the data to the sending out of the alarms and predicted values,” says Koen De Bauw. “Last but not least, ‘Yanomaly’ offers a lot more functionality and possibilities than we are actually using today. That’s not bad, because processes in the energy sector are also becoming increasingly complex. This tool is perfectly capable of growing along with our business, so that the investment will continue to pay off in the future”.

Who is Engie Laborelec ?

Since its foundation in 1962, ENGIE Laborelec (an entity of the ENGIE group) has evolved into a major centre of expertise and research in the field of electrical energy technology. Drawing on the skills of 280 engineers and specialist technicians, the company provides operational support, technical consultancy and applied R&D services across the entire electricity value chain: from generation, overtransmission and distribution, to storage and end-use. It has a particular focus on the energy transition and its three key concepts: decentralisation, decarbonisation and digitalisation. ENGIE Laborelec is a global player with activities in more than sixty countries, with headquarters in Belgium as well as offices in the Netherlands, Germany, Chile, Thailand and Abu Dhabi. Read more about Engie Laborelec.

Who is Yazzoom?

Since 2011, Yazzoom has been providing software and R&D services for analyzing data and improving business processes in various sectors. With the expertise in artificial intelligence, computer vision, software development and control technology of about 15 employees, it creates highly innovative solutions. In addition to advice and training, Yazzoom also creates software that can be adapted to the needs of the customer. In addition to its own software products ‘Yasense’ and ‘Yanomaly’, the company is also a distributor of the control technology software and services of PiControl Solutions LLC. Read more about Yazzoom.

Read more about this theme: