Anticipating maintenance problems with predictive analytics

By combining retrospective data analysis with predictive tools in advanced analytics platforms, process manufacturers can predict failures, enhance productivity, and establish optimal maintenance schedules.

Process engineering teams have in the past relied on stored data to improve operational efficiency, through preventive maintenance, failure mitigation strategies and process optimisation. These efforts emphasised monitoring and were built around available operational data — often recorded manually — to identify problems.

As time passed, these explorations progressively began using data stored in process historians and other databases to refine insights, leveraging diagnostic analytics to investigate issues and anomalies.

However, these procedures are reactive in nature, and the aim today is to move more towards proactive approaches that leverage historical data and context to drive more accurate process improvement decisions. Modern advanced analytics platforms are empowering manufacturers to implement proactive protocols that anticipate and address potential issues before they become failures, providing improved performance, enhanced operational efficiency, and increased uptime.

Today, advanced analytics can guide organisations through problem-solving journeys by examining historical, current and predicted time series data, which unveils insights that drive improved decision-making.

Types of analytics

The word ‘analytics’ is typically associated with IT software products, platforms and the cloud. We therefore need to qualify what we mean when we use the term.

For example, advanced analytics describes the use of statistics, machine learning and artificial intelligence in data analysis to glean insights. Other modifiers can be used to differentiate the analytics type based on utility and complexity.

‘Diagnostic analytics’, for example, describes the use of information from past events to respond to a given situation, investigating a raw set of historical data and applying statistical analysis to identify patterns and produce insights. In this case however, there is an unavoidable lag between the event or issue under analysis and the action taken to improve future performance, eliminating the ability to predict events before the fact.

Another type, ‘descriptive analytics’ also uses past data, summarising events via reports that can be more easily interpreted and learnt from. In this practice, the same lag occurs and it can be difficult to develop clear-cut answers on how and when to make decisions that will prevent future performance issues.

In contrast, ‘predictive analytics’ aims to deliver future predictions by using historical data to develop and train models that project future data, enabling teams to foresee plausible occurrences and informing them of actions they should take to drive desired outcomes.

Maintenance-focused advanced analytics

Understanding the historical and current behaviour of an industrial system is critical before attempting to predict future performance. This requires engineers to have access to real-time data from all relevant processes and databases. However, traditional data silos and disparate systems often hinder this access, creating challenges such as managing multiple logins and navigating different interfaces.

Advanced analytics platforms address these and other challenges by centralising data from various sources into a cloud-based platform. This eliminates the complexities of data connectivity and provides subject matter experts with streamlined tools for data cleansing and contextualisation. By unifying data in this way, experts can efficiently extract meaningful insights and build a comprehensive understanding of operational performance.

With live data connections established, organisations can leverage advanced analytics to generate valuable predictions, including those related to equipment maintenance. Predictive maintenance, a key application of this technology, aims to anticipate equipment failures or maintenance needs before they occur. By analysing historical data and identifying patterns, these models help improve quality, reliability and uptime by enabling proactive maintenance strategies.

The predictive capabilities of advanced analytics platforms stem from sophisticated algorithms that map organisational workflows and procedures to relevant laws, theorems or design principles. This mapping process enables creation of key performance indicators and enables accurate data extrapolation, extending the timeframe for insights and analysis. The following examples demonstrate how advanced analytics has empowered process manufacturers to optimise operational efficiency and transition from reactive to predictive maintenance approaches.

Example 1: Monitoring compressor health

Predictive maintenance strategies are frequently applied to the detection of compressor performance issues. Compressor failures can cause sudden and catastrophic shutdown or environmental safety concerns.

In one large manufacturing facility, data scientists leveraged machine learning algorithms to drill down to the root cause of compressor failure. By superimposing the algorithm on live data to identify signs of degrading performance, they were able to utilise their findings to create maintenance notifications ahead of expected failures, which provided insights on a visual interface for operators and process engineers.

These types of tools empower engineers to identify leading and lagging indicators of degrading compressor health and to continuously monitor variables, helping teams proactively identify risks and prioritise maintenance activities.

Example 2: Mitigating reactor failure

Collaboration between teams — such as process, maintenance and reliability — can be strengthened by leveraging built-in tools within advanced analytics platforms for sharing analyses and insights in easily digestible dashboards and reports.

One petrochemical and refining company was experiencing significant reactor shutdowns caused by a failing critical feed gas compressor on a polyethylene line. These failures had a high impact on production, preventing any way to immediately restart the process. Such unplanned reactor shutdowns were causing a minimum of four hours of downtime, costing the plant upward of US$200,000 with every incident. Previously, attempting to prevent such occurrences, the compressors were maintained on a preventive maintenance schedule, but this did not entirely prevent unplanned shutdowns.

Previous manual attempts to investigate and identify the safety interlock that prompted the shutdown failed to yield a root cause. A process engineer at the refinery therefore took an alternative approach, using an advanced analytics platform to rapidly locate the five most recent shutdowns and subsequent restarts — planned and unplanned — from decades of historical process data. With time-dissection tools, they focused on shutdown and startup time periods and overlaid all events, presenting abnormalities in the discharge pressure profile of the two most recent startups (Figure 1).

Figure 1: Facing a critical feed gas compressor failure, a petrochemical refinery used advanced analytics to pinpoint the five most recent shutdowns and restarts. Source: Seeq. For a larger image click here.

Upon further investigation, the engineer also identified early warning signs on the motor amperage signal. Without a method to view the startups back-to-back, the motor degradation had gone unnoticed by operations.

As a result of this root cause analysis, the process engineer implemented a monitoring solution to identify and flag future motor degradation to prevent similar unplanned shutdowns. When an out-of-tolerance value appears, the compressor motor is now immediately added to the maintenance work list for the next planned shutdown — a proactive maintenance approach that is expected to eliminate unplanned shutdowns due to this failure mode.

Example 3: Predicting valve erosion

Valves play critical roles in almost all process plants. Keeping valves in top condition is essential for maintaining efficient operations, but condition monitoring can be tricky based solely on observation. Advanced analytics solutions often provide low-effort, high-return opportunities to monitor valve conditions and reduce unexpected failures, simultaneously protecting adjacent process equipment and devices.

An oil and gas producer operating more than 50 well pads — each with a gathering system containing a critical flow control valve — was experiencing frequent valve failures. Each failure occurrence rendered the pad inoperable for days until repairs could be made. These failures were typically caused by sand erosion, and the producer had no methodology in place to determine early warning signs other than exhaustive manual inspections, which required complete shutdown and became increasingly cost-prohibitive as the asset base grew.

To solve this problem using advanced analytics, subject matter experts leveraged both real-time and historical process data to calculate a metric indicating progressive erosion in the valve seat, and were able to establish an indicator to predict future failures. The team leveraged first principles to produce this metric, which was used as the basis for a predictive model to approximate time to failure.

Figure 2: An oil and gas producer leveraged predictive analytics to anticipate valve failures, enabling proactive maintenance and minimising operational disruptions. Source: Seeq. For a larger image click here.

This analysis was scaled to all well pads by leveraging historian hierarchies imported into the analytics platform through its native connectors. By deploying an advanced analytics platform to monitor process conditions, the oil and gas producer was now able to predict erosion progression and approximate the time to valve failure, enabling the maintenance team to prioritise service, significantly reducing downtime and operating expenditures.

Making advanced analytics work

The future of process manufacturing hinges on in-depth knowledge of past equipment behaviour. Advanced analytics platforms combine retrospective and predictive analytics, empowering process experts and data analysts to efficiently construct robust models, forecasting maintenance needs and illuminating paths to mitigate risk.

Armed with this potent digital arsenal, process manufacturers build better models, providing plant insights and projecting issues prior to failure so personnel can optimise maintenance schedules and prevent costly downtime.

Top image credit: iStock.com/Lahiru Lakmal