Compressors, pumps and turbines, heat exchangers and valves – each individual component in an industrial plant has its specific function. Just like in a symphony orchestra, they have to work perfectly in unison for harmonious results. For many of Linde Engineering’s plants, a central location assumes the function of the “conductor” – the Linde Remote Operation Centre (ROC). Five worldwide locations manage around 1,000 plants in total. During remote operation, huge volumes of data flow – collected by hundreds of thousands of sensors that monitor the system’s operation. Linde’s digitalisation team has developed a program that uses this data trove. It allows engineers to detect faults or failures before they happen. With the predictive maintenance approach, damage and the resulting expensive downtimes of industrial plants can be avoided - thus securing operating revenues in the millions.
Remote Operations Center (ROC) Kuala Lumpur, Malaysia
Analysing data streams, detecting damage early
At the ROCs, the engineers continuously monitor the data streams that come from the 500,000 sensors and are stored in a central repository. Each plant has an average of 500 sensors that constantly collect health check data. They detect physical data including pressure, temperature, component vibrations and the flow rates of gases and liquids. Any deviations between the values read by the sensors and the target values can point to a future malfunction. However, the diagnosis is almost never clear-cut. The reason is that all of parts of the plant are networked and every deviation can also affect other components. “We receive several gigabytes of sensor data every day,” points out Linde engineer Christoph Heckmann. “There is no way that anyone could sift through all of that by themselves.” Although this job was too big for one person, it wasn’t too big for Linde’s digitalisation team. It joined forces with the Karlsruhe-based data analysts from the start-up anacision to ”train” computers to analyse the data. Together, they analysed the point at which a plant’s health starts to deteriorate and fed these insights into an algorithm. This is now at the heart of Linde’s Predictive Maintenance project.
Saving costs through optimal maintenance cycles
The algorithm bundles the know-how that Linde Engineering has gained in plant design and construction and the hands-on insights that Linde Gas has gained in plant operations. Sophisticated mathematics provide the glue. “Manufacturers of individual parts often have zero experience on the operational side,” according to Heckmann. “Rough estimates are all they can offer for recommended maintenance intervals.” Sometimes these recommendations are way off the mark. And that can lead to parts being replaced even though they were still in good working order and posed no immediate risk. Conversely, the early replacement of a failing component can help to avoid an operational outage. Both scenarios lead to costs that can be avoided with a better understanding of the equipment. Maximising plant availability is the ultimate aim.
In the South-East Asia region (RSE) alone, Linde is hoping to achieve annual savings of four to five million euros by optimising maintenance cycles. The pilot phase of the Predictive Maintenance project is already underway at the Kuala Lumpur ROC. The first component implemented in the algorithm by the programmers was the main air compressor, a key element of every industrial gas plant. “When it goes down, everything comes to a standstill,” emphasises Heckmann, comparing it to a car’s engine. Other plant components will be added on a phased basis. In due course, the program will be extended from Malaysia to other regions.
Regional Operations Centre, LLH Taichung, Taiwan
Intelligent industrial plants thanks to data analysis
“Going forward, we are aiming to monitor the entire plant and deliver predictive maintenance using our new system,” explains Heckmann, outlining the aim of Linde’s engineers. He is convinced that “in future, we will only be able to sell plants that have smart data analytics included in the package.” Linde’s engineers have long solved the challenge of sourcing the data needed to power this analytics solution. “Data from half a million sensors gives Linde a considerable head-start in this developing these capabilities,” affirms Heckmann. The company is sitting on a gold mine of data and the team of programmers and data analysts supporting Heckmann have only just started to extract value from it.