AI-based maintenances improve pump uptime
Data-driven information supports advanced predictive maintenance.
Industrie 4.0 and digitalization offer optimization opportunities for operations and maintenance functions. Innovations such as digitalization, artificial intelligence (AI), machine learning (ML), neural networks, and cloud computing have raised the capability to collect, analyze and trend equipment condition/health in real time.
With these advanced monitoring equipment and analytics methods, reliability engineers can step up predictive maintenance (PdM) programs to achieve profitability goals, including:
- Optimizing critical assets’ service life
- Minimizing unscheduled downtime
- Controlling maintenance costs
- Improving plant safety and operations.
Maintenance can be a profit center
Energy, power and chemical/petrochemical processing plants are very complex and complicated facilities. Numerous equipment and infrastructure items manage, contain and store feedstocks, process fluids and gases. Decreasing process unit and plant downtime due to unavailable equipment and systems are central to improving company profits.
Mechanical failure is the leading cause of processing industry accidents (Figure 1), while equipment failures result in 36% of unscheduled plant shutdowns (Figure 2). Creating better maintenance and planning programs increases operational excellence and facility uptime.
‘Smart’ maintenance profitable
According to a recent McKinsey report, PdM can increase asset availability (either process unit or equipment) by 5% to 15%. Optimized PdM activities lengthen the service life of key assets by 20% to 40%. More importantly, PdM effectively can reduce maintenance costs by 18% to 25%. Improved monitoring and early proactive maintenance significantly reduce repair and replacement costs for key processing equipment and minimizes unscheduled downtime and lost productivity. In addition, unexpected equipment failures may result in losses greater than the replacement value of the asset.
In general, rotating and reciprocating equipment have the highest failure rates (Figure 3). Vibration problems are predominant root causes for rotating equipment failures, especially pumps. All rotating equipment vibrates, however, the changes in vibration levels over time are indicators of possible problems. In the hydrocarbon processing industry, about 7% of the pumps in use consume 60% of the money spent on pump maintenance and repair. Finding and addressing the root causes for vibration or temperature changes supersedes just treating the symptoms.
To avoid repeat failure, pump owners must push routine maintenance practices to a superior level. Increased use of smart manufacturing strategies and cloud computing can raise the integrity of PdM activities.
Not a new concept
Since the 1970s, maintenance and reliability engineers installed stress (piezoelectric) sensors to monitor and detect performance issues on pumps and motors. Unfortunately, these early methods encountered trending and continuity problems with data collection. These vibration sensors often operated at different frequencies and amplitudes and had their own baseline signatures.
Deciphering collected sensor data into usable information required interpretation by data scientists. Stress sensors often were lost or removed by routine maintenance actions. Early sensors were connected physically to minicomputers or terminals by wires. Contaminants such as dirt and lubrication oil degraded sensor signals. Reliability and maintenance engineers found that early vibration-monitoring methods did not yield desired results.
Improvements in minicomputers, terminals and hand-held sensors improved equipment monitoring programs. However, real-time information and data connectivity remained limited due to computer hardware and software capabilities. Converting vibration sensor signals into useable information remained a tedious task. In addition, data trends and information were siloed in databases and not easily shared among users.
Remote monitoring methods were incorporated in preventive maintenance programs. The amount of data collected was never the issue. In some cases, too much data decreased the ability to find key information on equipment health. The long-standing problem remained understanding what the data were indicating about an asset’s condition. Simply put, you can’t effectively correct what you don’t understand.
Vibration monitoring methods still struggle to provide and convert collected data into reliable real-time information. Too often, routine and preventive maintenance programs discover deteriorating conditions of rotating equipment after significant damage has occurred.
Over the last 10 years, the use of wireless technologies, cloud computing, smart-field devices and AI enable the management of plant assets through advanced PdM programs. While preventive maintenance is done on the manufacturer’s recommended schedule, PdM pushes maintenance activities to the next level. In PdM, real-time process and equipment data build trends and histories that be used to forecast changes within process equipment. Improving equipment availability and process uptimes through enhanced reliability/maintenance programs, such as advanced PdM, can increase operational excellence and plant safety.
Advanced PdM analytics
To be fully effective, PdM programs require robust and valid data and the analytics to develop information-driven decisions. Recent advancements in AI and ML enable analyzing and converting huge volumes of collected data into patterns. To monitor rotating and reciprocating equipment, advanced vibration sensors use cloud computing to upload real-time data in various formats. Innovative AI and ML algorithms, built on a combination of software and neural networks, convert and analyze wireless sensor data.
This information generates trends that identify normal and unhealthy operations. More importantly, the AI algorithms “learn” from the transmitted vibration data and discern between “normal” or unacceptable signals.
Using AI-based predictive analytics, ML and neural networks, correlations concerning the performance of critical equipment are possible. These validated analytics are instrumental in identifying true root causes for performance deviations of rotating and reciprocating equipment. Early fault identification enables optimum corrective actions to be selected before substantial asset damage or failure occurs, thus minimizing repair costs, reducing unscheduled downtime and ensuring safe operations.
Deteriorating conditions of pumps and compressors are not observed easily through visual or normal health checks. AI and ML algorithms identify patterns from the histories and detect performance decline as proven by deviations in asset trends. Equipment performance problems are identified much sooner than through traditional preventive maintenance methods. System-generated alarms alert maintenance engineers to conduct further investigations.
Also, AI-based predictive analytics go beyond failure notification. They use data and operating histories to estimate the remaining useful life (RUL) of failing equipment or a component item. With RUL, maintenance and reliability engineers have complete information to plan repair and replacement actions that have the least impact on process uptime.
Visualization of data
Data without refinement have limited value. The basis of PdM is visualizing the data. Advanced PdM uses AI-based analytics and neural networks to distill collected data into usable information. They usually include dashboards that enable users to quickly survey equipment condition data and review trends. With such graphics, engineers can interpret equipment/process unit health easily and make more informed, data-driven decisions. In addition, RUL estimates are combined into the graphics for centralized information.
Secure wireless technology and mobile apps connect advanced sensors to the cloud for analysis by AI and ML software. Fully applying Industrial Internet of Things (IIoT) and cloud computing, maintenance and reliability professionals continuously can review the health of critical process equipment. The ability to forecast the RUL and “time to failure” is invaluable. With such information, maintenance and operations groups can plan repair actions rather than react to an emergency shutdown or unscheduled outage.
Case history: L&T Nabha Power plant
The L&T Nabha Power facility is the company’s first supercritical, coal-fired power plant and is one of the most efficient power-generation facilities in India. This facility operates two 700-MW supercritical thermal power units and is the major electricity provider to the state of Punjab in northern India. As the chief regional energy provider, reliability of the Nabha power plant is critical. Unplanned maintenance and shutdowns of this facility have dramatic and adverse effects on the productivity and profitability of regional businesses and residential customers. Unfortunately, this power plant experienced three unplanned downtimes in one year due to critical-service pump failures.
In power generation, pumps are key processing equipment. The condensate cooling-water pump is one of the critical-service pumps to maintain steady-state operations of the facility (Figure 4). It is a horizontal vane pump operating at up to 1,650m3/hour with a discharge of 9 MPa (62 psi) at 986 rpm. Each day that this pump is offline costs the power plant up to $250,000 in lost revenue. Unplanned maintenance and failure of this pump can incur repair costs exceeding tens of thousands of dollars.
The facility’s condensate cooling-water pump had chronic unscheduled downtime due to bearing failures from cavitation events. As the primary electric power provider for the region, reliability of the condensate cooling-water pump was a top priority at the facility.
To resolve failure conditions and increase unit uptime, plant engineers elected to install a real-time vibration monitoring and advanced AI-based predictive analytics solution on the condensate cooling-water pump. The new monitoring strategy focused on early fault detection of the pump and its components. Besides fault detection, this monitoring solution included AI-based algorithms to provide reliable RUL estimates before service interruptions occurred. To solve the bearing failure problems for this condensate cooling-water pump, several advanced wireless sensors were installed to monitor:
- Non-drive-side bearing, electric motor
- Drive-side bearing, electric motor
- Drive-side bearing, pump
- Non-drive-side bearing, pump.
The monitoring system used secure Wi-Fi-enabled sensors to collect and upload vibration data continuously via the cloud. Cloud-based AI algorithms analyzed and tended the collected data.
Approximately six weeks after the installation of the advanced sensors and analytics system, the new monitoring program alerted maintenance staff that a vane fault had developed. It was causing cavitation problems for the condensate cooling-water pump. The plant’s maintenance staff verified the fault with a hand-held vibration monitor and did a partial disassembly to visually confirm damage to the pump vanes. A temporary repair of the damaged vanes was done before putting the pump back in service.
The advanced AI-based PdM system estimated a RUL of 25 days before total failure. This was sufficient time to schedule the pump replacement during an already planned maintenance outage. Applying AI-based predictive capabilities and advanced vibration monitoring, L&T Nabha Power avoided a serious pump failure and unplanned downtime. Early intervention reduced much needed repairs and minimized interruptions to facility operations.
Not just black boxes
IIoT, cloud computing and wireless technologies support AI-based data analytics as part of an advanced PdM program. Fully applying AI and ML methods, engineers can detect anomalies or faults in critical-service equipment well in advance of failure mode. Advanced wireless vibration/acoustic sensors support PdM programs in collection and uploading of real-time data.
With AI, ML and neural-network algorithms, advanced analytics develop historical trends of monitored equipment or components. With a complete operating history, AI-based analytics identify changes in the trending data and estimate the RUL of the monitored asset. With the RUL, maintenance and reliability engineers can take corrective action well before failure and focus on preserving the asset and maintaining safe operations. Advanced PdM programs support better results-driven maintenance plans that improve operations uptime and safety.