RECENT conversations in the media and industry press discuss the potential of big data, the Industrial Internet of Things and Industry 4.0 to transform performance in the process industries1,2.
There is a great opportunity to apply these technologies to update the traditional maintenance methodologies used in process industries. However, true ‘smart maintenance’ means deploying these innovations together with traditional approaches where appropriate, within a broader maintenance system that in turn supports overall operational and business objectives (see Figure 1).
This article aims to explain some of these smart maintenance technologies, and how to implement them within an overall maintenance strategy to give the uptime, speed and yield that process plants need without compromising safety and security.
Due to the inherent limitations in failure predictability and human capability associated with traditional maintenance techniques, there are several factors encouraging operating companies in the process industries to adopt smarter maintenance practices.
Loss of experience – fragmentation of operating companies and an ageing workforce means less on-site skill and experience within operations, maintenance and engineering teams.
Market pressures – competitive challenges mean an ever-continuing need to improve reliability. Manufacturing operations, underpinned by the maintenance organisation must achieve better product quality performance and conformance, delivery to shorter lead times, and greater flexibility and dependability.
Cost pressures – continuing cost and profitability challenges mean that producers are looking for ways to increase plant availability while decreasing maintenance spend. For example, the ‘lower for longer’ oil price has made the business case for continuous offshore presence of vendor-supplied maintenance technicians on production platforms unviable, and so operating companies are trying to achieve round-the-clock support ‘from the beach’.
Ageing assets – maintenance gets harder as assets age. Consequently, companies are looking for ways to understand their assets better and proactively look after them.
A key component of a site’s overall maintenance strategy (see Figure 2) is the maintenance policy applied to each equipment item. A policy is made up of one or more maintenance routines3. There are three approaches:
Figure 3 shows the principle behind condition-based maintenance using a P-F Curve4. The X and Y axes represent time and condition of the asset, respectively. F is the point of failure. P represents the point at which a condition monitoring technique can begin to detect potential failure. The technique could detect variations in product quality, process or equipment parameters. Provided it gives a sufficient interval between P and F to perform the required maintenance action then condition-based maintenance avoids costs from:
Although numerous techniques have been used in industry for condition-based maintenance, they often had drawbacks that limited their implementation to the most critical of equipment such as aircraft gas turbine engines. In particular, it was expensive to achieve a measurement sensitivity and sample rate to give a useful P-F interval. Fortunately, several recent technological developments are making condition-based maintenance an appropriate approach for a wider range of process plant equipment. Not only has it become more cost effective to get a useable P-F interval, these advances enable condition-based approaches to go beyond descriptive monitoring of condition to encompass capabilities that are:
Several advances in computational and connectivity technologies make diagnostic, predictive and prescriptive capabilities possible.
Better and cheaper condition monitoring techniques are emerging because of the continual increase in the amount of computing power available for a given size, cost and power consumption. Moreover, accelerometers, gyroscopes, GPS and camera modules originally designed for mobile devices can be bought cost-effectively ‘off the shelf’ for condition monitoring. An example is ABB's Ability Smart Sensor, with the size and cost of a budget smartphone, but a much better battery life! It is retro-fitted to motors, pumps and mounted on bearings to monitor vibration, temperature and other parameters.
High performance computing power makes it possible to use machine learning algorithms to automatically monitor and react to asset condition data, such as vibration traces from rotating machinery. These algorithms do not use a predefined set of rules, as in traditional software programming. Instead, being a type of artificial intelligence (AI) technique, they are self-learning. They infer rules by performing a series of trials on a set of training data and thus construct their own model of the world. Subsequent ‘live’ data refines the model and improves its predictive powers, as it learns more about the equipment being monitored. Large operating companies can store petabytes of asset monitoring data ready to analyse in ‘data lakes’5. However, even these are dwarfed by the quantity of data produced by a modern distributed control system (DCS) on a large asset. Fortunately, a technique called stream analytics summarises the flow of data by doing real-time processing of product, process and equipment data to pick out key information such as events and trends.
Connectivity enablers make it possible to collect and share enormous amounts of life cycle information about process equipment. This is the central idea within the “Industry 4.0” concept. Data about product design from OEMs (original equipment manufacturers), system design from systems integrators and overall plant design from EPC (engineering, procurement and construction) companies can be combined to create a ‘digital twin’. This is a virtual model of the asset that exists in parallel with the physical asset, from its specification to its disposal. Maintenance teams can use the information from the digital twin to devise effective maintenance routines. When spatial and photographic data are incorporated into the twin, a complete “Google Street View” can be built up of the asset. This information makes planning, executing and documenting projects, inspections and maintenance activities more effective.
When this design information is combined with operational asset and process data from condition monitoring equipment and SCADA (supervisory control and data acquisition) systems, it becomes possible to analyse anomalies with physical equation-based models rather than statistical ‘black box’ models. As a result it is possible to accurately diagnose problems and precisely predict failures, using information from product manuals to prescribe appropriate maintenance actions.
The other aspect of connectivity is the “Internet of Things’, which gives the ability to collect operational data from thousands of devices in real time via wired or wireless industrial networks, meaning that asset condition information can now be obtained with short sampling intervals and analysed without delay. In contrast, many traditional condition monitoring techniques rely on manual data sampling, which leads to schedule slips and delays in compiling results. Improved connectivity also enables asset health information to be shared securely beyond a plant’s on-site network to other locations for analysis or reporting in corporate-level dashboards. The cloud refers to managed, offsite computing facilities that are connected over the internet to local devices via web browsers. Generally, they are ‘remote clouds’ hosted on external systems, such as Microsoft Azure and Amazon Web Services. If an industrial facility produces too much data to upload to the cloud, the condition monitoring system is built into the process control loops, with some local data processing or ‘edge computing’. The company can also restrict the data that leaves the site, by creating its own ‘local cloud’.
Connecting process plants to the cloud also allows for ‘collaborative operations’. For example, in case of a lack of on-site capability to interpret and act on asset condition data on-site, you can give access to specialists from corporate engineering centres, OEMs or third-party service companies. Moreover, external maintenance engineers and service technicians can be given access to the ‘digital twin’ data on the equipment before arriving on-site, allowing for a better service.
Compiling data from different plants allows for ‘fleet analytics’, where the operating company OEM merges the data coming from similar devices on different facilities to give a much broader range of operating windows, failure modes and effects than by looking at a single device operating in isolation. This then allows a more accurate P-F curve to be identified.
The drive, motor and driven machinery for critical equipment such as a gas compressor train (see Figure 4) can now all be comprehensively monitored. The machine control system, drive CPU and the machine monitoring system (see Figure 5), can collect process, electrical, mechanical and machine operating data. Wireless monitoring can also be used. This data is used to monitor failure modes related to the compressor. One mode is fouling, the formation of polymers on the compressor internals. Early detection of this prevents an increased pressure drop throughout the machine and associated increased energy requirement. Proper monitoring of stiction on the anti-surge valve by analysing the response time of the valve allows evaluation of the valve’s capability to protect the compressor from a fast surge event. When the response drops beyond a critical level, maintenance action is initiated.
Despite the potential of smart maintenance technologies, it is still often appropriate to do scheduled or breakdown maintenance. Plant management needs to develop a strategy to drive decisions about application of each maintenance type. Implemented technologies need to be appropriate to the maintenance organisation’s maturity; predictive maintenance approaches are unlikely to work effectively with an uncontrollable scheduled maintenance backlog. The maintenance strategy also needs to be consistent with the operations and wider business strategy, with a business case for investing in any technology, and prioritised investment in the most beneficial areas.
Key drivers of the maintenance policy decision are the business and safety criticality of each equipment item in the asset register. Once the high priority items requiring condition-based maintenance are identified, a failure modes and effects analysis (FMEA) can be done to identify which equipment or process performance parameters can be monitored to detect impending failure, along with corrective maintenance actions.
For smart maintenance technology to make a difference in plant performance, it needs to result in service action. The outer loop in Figure 6 must be closed. Operators and maintenance technicians need to first trust the technology and feel compelled to act and respond appropriately based on what the technology is telling them. They therefore must be involved in the implementation of predictive maintenance technology. Sufficient technical effort is also needed to document maintenance routines, and ensure the workforce has the competence to implement them. The technology can only detect those failure modes that it has been trained or programmed to look for. In many cases the 'smartest' routine is a regular low-tech ‘look, listen and feel’ round.
Sustaining the benefits of investing in condition-based maintenance tools means embedding them into the management system and organisational culture. People should use smart maintenance systems as learning tools alongside root cause analysis (RCA) of failures (of machines, people and systems). Both should be used to continually update maintenance policies, plans and procedures. This is especially important with ageing assets as new failure modes emerge.
1. "The Chemical Sector and its Digital Journey", The Chemical Engineer, May 2017.
2. Big Oil Harnesses Power of Data Analysis to Ensure Survival, FT 14 June 2018.
3. Ron Moore, Making Common Sense Common Practice, 3rd Edition, Elsevier, 2004.
4. John Moubray, Reliability-Centred Maintenance, 2nd Edition, Butterworth Heinemann, 1997.
5. Looney, Bernard, Petroleum Review, Energy Institute, Dec 2017/Jan 2018.