REFINING INDIA 2025

refining india 2025

Automated anomaly detection and lifetime prediction for reciprocating compressors

Gunther Machu HOERBIGER

Reciprocating compressors are the work- horses in the oil and gas industry, enabling the reliable production of process gas, various refined fuels, and chemicals. So far, maintenance schedules and trouble- shooting have been based on the decades of experience and know-how of individ- ual experts on these machines, as many of these units are custom-built and lack adequate instrumentation to capture pro- cess signals. Now, more and more of these experts are retiring, and their expertise is difficult to replace. Here, modern analytics in conjunction with advanced sensors and data acqui- sition come into play. By combining ded- icated physics-based models of the compressor and its components, such as compressor valves, rings and packings, with machine learning algorithms, it is pos- sible to detect anomalies at an early stage. This establishes a solid basis for predicting remaining useful life. The key is to display the results in an intuitive, easy-to-under- stand, and actionable way. Compressors Reciprocating compressors are used in many processes of the oil and gas indus- try, mainly due to their ability to efficiently and reliably compress even light gases like hydrogen to very high pressures. The key components in recips that cause 80% of unscheduled breakdowns are compres- sor valves, off-design process conditions, rings and packings, and cylinder lubrica- tion systems.¹ Over the past decades, the reliability and performance of these key components have improved significantly. It is not uncommon for valves to run 16-32,000 hours without failure. Hence, this increase in component reliability is directly translated into com- pressor uptime and availability. There is, however, very limited insight on the health, performance, and remain- ing useful life of these components while the compressor is in operation. Hence, most operators still adhere to scheduled maintenance, mostly following the origi- nal equipment manufacturer’s recommen- dations. They are therefore unable to reap the full benefits of extended maintenance intervals made possible by the improved components. Here, predictive maintenance comes into play – the ability to identify early anomalies and predict the remaining useful life of the compressor components, allowing for con- dition-based maintenance and thus a sig- nificant reduction in maintenance costs (not to mention loss of production due to unscheduled breakdown). This is exactly what Hoerbiger‘s Vistra ® Predict, addresses, answering three press- ing questions of operators:  Are my compressor and components running healthy? v Can my compressor be optimised, hence adapted to new operating requirements?

DID you Know? VISTRA Predict displays component

anomalies in a simple and

actionable way, showing findings and location on the unit

way on the front end, showing the loca- tion on the throw and the result ( Figure 1 ). Of course, the data evidence can also be accessed by those reliability engineers who want to dig deeper. The evidence that Vistra Predict analyt- ics found in the data was a steeper com- pression and flatter re-expansion curve in the indicator pressure vs crank angle reading of that compression chamber ( Figure 2 ). The physics-based anomalies detected also serve as the basis for life- time prediction: tracking those anomalies over time leads to a conditional probabil- ity of failure as well as a remaining useful life forecast. Conclusion This article has demonstrated that field instruments on reciprocating compressors can be used to provide intuitive insights into component anomalies and remaining useful life in Vistra Predict without requir- ing expertise in compressors. Hoerbiger is now launching this combi- nation of software and field instruments on a commercial basis for the benefit of users of reciprocating compressors. Literature 1 André Eijk, European Forum of Reciprocating Compressors (EFRC) Reliability Study 2019 .

Figure 1 Intuitive overview of the compressor and the anomalies found on a component level

data to mathematically reconstruct failure modes and root causes in real time. w Machine learning from historic com- pressor data to predict future component failures. Results Hoerbiger installed the first pilot instal- lation earlier this year in Kota, India, at a fertiliser plant on a six-cylinder hydrogen compressor with 4.5 MW power consump- tion. Indicator pressure sensors, cylinder, crosshead and frame vibration probes, as well as rod drop sensors, were installed on each throw. In addition, valve cap temper- ature sensors were used to evaluate the automated analytic results. The Predict DC system acts as a black box; only a power supply is needed from the customer, eliminating any interference with existing systems. All sensor data are sent continuously to the Hoerbiger cloud via a 4G cellular modem. After start-up, Vistra Predict analytics immediately identified a leaking discharge valve on one throw of the compressor, causing production loss. Based on the data insight, the customer has enough lead time to correct this issue at the next planned stop. The cloud analytics software displays this anomaly in a simple and intuitive

w What is the remaining useful life of my compressor? Vistra Predict is cloud-based software that runs 24/7/365 and analyses data from compressors in an automated way. Methodology Hoerbiger’s system architecture for auto- mated anomaly detection and lifetime pre- diction consists of the following: • Proprietary analytics with a front end customer-facing user interface. • Cloud repository for customer data – on Hoerbiger’s MS Azure cloud. • A cybersecure, automated file-based data transfer where customers push com- pressor data files to Hoerbiger’s cloud repository in a continuous way. • Instrumentation for the compressor, including sensors, data collection, and an industrial edge device for smart preproc- essing of the sensor data – Vistra Predict DC – data collection. The data files are analysed continuously, and results are accessible for customers on the intuitive front end, using the follow- ing analytics:  A physics-based virtual compressor model (digital twin) that includes all com- pressor components, using 130 years of Hoerbiger’s design expertise. v Reconstruction methods using sensor

Contact: gunther.machu@hoerbiger.com

Pressure-Crank angle (Pa)

28/03/2025 07:30:48

33,000,000

32,000,000

31,000,000

30,000,000

29,000,000

28,000,000

0

84

168

252

336

Throw 2 HE Pressure

Throw 2 CE Pressure

Throw 4 CE Pressure

Throw 4 HE Pressure

Figure 2 Data analysed by Vistra Predict, and field installation of the sensors

23

Powered by