Predictive maintenance is an equipment maintenance policy for the monitoring of an organization’s physical assets to detect potential problems in the future and perform corrective actions before components malfunction or fail. The types of assets typically monitored include large and valuable equipment like field instrumentation and machines, and small but essential equipment components like valves and fan belts.
A predictive maintenance system is the practical application of a predictive maintenance policy. A predictive maintenance system consists of sensors attached to field hardware, the means to integrate with industrial control systems to gather sensor data and perform corrective actions, and predictive analytics software to analyze data and make predictions about potential future equipment failures. Central to a predictive maintenance system is a control hub via which tasks for maintenance crews are scheduled.
Before organizations implement a predictive maintenance policy, they must design a predictive maintenance program (plan of action). A predictive maintenance program describes the assets that will be maintained and identifies potential points of failure, called failure modes, under different operating conditions.
Failure modes are the specific ways components can potentially fail. Examples of failure modes are human error, corrosion, erosion, wear and tear, and metal fatigue. Failure modes may result in malfunctions like misaligned fan belts, pump seal failures, lubrication defects, and empty fuel tanks. Components may have multiple potential failure modes.
The predictive maintenance model for asset maintenance evolved in response to limitations and disadvantages of preventive and reactive maintenance models. For example, preventive maintenance is labor-intensive and can result in over-maintenance, and reactive maintenance is time consuming and can result in sporadic downtimes.
An organization’s asset maintenance strategy usually combines different types of asset maintenance approaches to address different types of problems. For example, reactive maintenance programs provide solutions for unexpected equipment failures. The goal of a predictive maintenance system on the other hand is to predict when equipment malfunction could occur and prevent failures through scheduled, corrective maintenance tasks.
Limitations of the traditional predictive maintenance model for asset maintenance has resulted in more organizations adopting a hybrid approach, combining condition monitoring, IoT-based predictive maintenance, and prescriptive maintenance techniques.
Predictive maintenance aims to perform corrective maintenance before equipment fails whereas breakdown maintenance is performed after equipment fails.
Examples of breakdown maintenance include reactive or corrective maintenance, run-to-failure (RTF) maintenance, and unplanned downtime.
When an asset has unexpectedly failed, insights about the cause, severity, and likelihood of the failure happening in the future provide input for predictive analytics algorithms that predict the likelihood of future equipment failures.
The upfront cost of implementing breakdown maintenance programs is low but the cost of repairing assets in breakdown maintenance programs is high because parts may need to be replaced instead of repaired and may not be readily available, which may result in long, and unplanned for, periods of downtime. Breakdown maintenance is a kind of “Plan B” approach.
Like predictive maintenance, preventive or time-based maintenance reduces the occurrence and impact of equipment failures.
Predictive maintenance involves performing maintenance tasks depending on the condition of components.
Preventive maintenance involves scheduling regular maintenance tasks that may be unnecessary, like replacing a part based on a manufacturer’s estimate of the part’s projected lifespan. Preventive maintenance is important in mission-critical and safety systems.
The terms condition-based maintenance (CBM) and predictive maintenance are sometimes used interchangeably but they are slightly different.
Both types involve the regular monitoring of sensor data to identify a potential fault or problem with a component.
Predictive maintenance techniques use formulas that combine real-time sensor data and historical data to make predictions about future maintenance requirements. For example, predictive analytic algorithms could estimate how long it will take before a tank needs refueling and schedule a task for maintenance crews to do so in the future.
CBM relies on real-time measurements to take action when sensor data is not within range of its optimal value, for example the volume of fuel in a tank is too low. In this scenario, maintenance crews may not have the resources to perform the refueling operation at that point in time, leaving a machine temporarily inactive.
Used together, predictive maintenance and CBM can help organizations to monitor equipment for current and future malfunctions.
Prescriptive maintenance models go a step further than traditional predictive maintenance models. Predictive maintenance systems predict potential equipment failures and suggest ways to delay or mitigate them in the future. For example, outcomes of a predictive maintenance program may highlight that a conveyer belt is likely to break down within the next few months.
With a prescriptive maintenance approach, outcomes would include information about how the conveyor belt could be operated to increase its lifespan, for example by reducing its speed, and reduce the inevitable downtime for maintenance crews to replace parts.
Because of the cost of implementing predictive maintenance solutions, traditional predictive systems focused on an organization’s most critical and valuable assets. The predictive maintenance of less critical and less valuable equipment usually involved irregular scheduled and random checks by maintenance crews.
In modern predictive maintenance systems, IoT technologies like multiple wireless connectivity options, low-cost micro electro-mechanical system (MEMS) sensors, integration with external systems, cloud computing, increased cloud-based storage capacity, and artificial intelligence make the reliable predictive maintenance of multiple assets across large production networks possible.
Predictive maintenance systems are especially important for the safe and reliable operation of assets in critical systems like oil & gas, mining, aviation, and industrial manufacturing, and at nuclear power plants and utility suppliers.
Predictive maintenance systems are commonly used:
- In industries that have valuable assets that are expensive to repair or replace, or require specialist skills to maintain, for example on space stations
- Where failure modes like electrical overloads can be reliably predicted
- Where asset failures like utility outages could affect service delivery
- In retrofitted environments
Some use cases for predictive maintenance systems are:
- Measuring the temperature of foods at food production facilities to reduce spoilage
- Tracking the condition of vehicles in the supply chain industry to avoid breakdowns
- Monitoring leaks on oil rigs
- Providing statistics to insurance companies about the likelihood of breakdowns of specific vehicle models
- Scheduling building and office maintenance
- Product quality management
- Predicting future requirements for network expansion
Predictive maintenance helps organizations to:
- comply with safety regulations
- take preemptive corrective actions on malfunctioning equipment
- reduce operational costs
- improve the health and performance of valuable assets
- reduce human error in guessing when a piece of equipment must be maintained
- minimize system downtime
- reduce the wastage of resources
- plan maintenance around their production schedules
Condition-monitoring sensors monitor the health, state, and behavior of components by gathering information about variables like temperature, corrosion, pressure, vibration, noise, and volume. Variables are condition indicators.
When designing predictive maintenance programs, data scientists specify what condition indicators are most useful to distinguish between the normal and faulty operation of a particular component.
Tools like infrared, oil, vibration, motor circuit, laser-shift alignment, and ultrasonic analyzers measure condition indicators from different types of sensors:
- Infrared analysis can determine the operating temperature of mechanical and electrical components.
- Oil analysis can detect the presence of contaminants that affect the normal operation of a machine.
- Vibration analysis allows technicians and operators to identify malfunctioning components from the different vibrations they emit due to wear and tear.
- Motor circuit analysis detects faults in a motor’s electrical circuit.
- Laser-shift alignment tools test the alignment of components like drive shafts.
- Leaks can be detected using ultrasonic analysis.
In modern predictive maintenance systems, machine learning (ML) models use historical sensor data and information about an asset’s performance to establish patterns and baselines of normal behavior. These models then look for deviations and anomalies in the real-time performance of assets and make predictions about if and when a component might fail.
ML algorithms use data not only from sensors but also from external sources, for example information from a field operator’s report, historical data from an enterprise resource planning (ERP) solution, or situational data like adverse weather conditions that may have an impact on the operation of specific components.
Work orders provide instructions for maintenance crews to repair components. Remedial tasks can be performed remotely, for example remotely closing a valve, or by human intervention, for example physically replacing a part.
Work orders are created via a CMMS, either automatically or manually. CMMSs also store historical data about asset performance and help to automate the scheduling of routine and corrective maintenance tasks. They provide a central hub to organize workflows and store information from external applications, like inventory and workforce management systems.
Typically, creating a predictive maintenance program has the following steps:
First, management and IT teams identify critical and valuable assets and document the optimal operation of these assets, including the range of desired values for specific parameters, like the position of blades on a wind turbine in specific weather conditions. This information provides baseline measurements for the desired operation of an asset.
Second, a CMMS database is populated with information about each asset. Historical records about asset maintenance, insights from maintenance and operational staff, and equipment information from manufacturers provide valuable insights into potential failure modes.
Third, system analysts use failure mode and effects analysis (FMEA) to identify the possible reasons system components could fail, the possibility of occurring failures (from extremely likely to unlikely), and what the consequences could be (from risk to life to slight equipment damage). Work orders are prioritized according to these rankings.
Fourth, technicians install condition-monitoring equipment like sensors and PLCs.
Fifth, data scientists create predictive algorithms to assess the real-time behavior and state of an asset against its baseline operation. The workflow involved in creating a predictive algorithm is to collect sensor data, preprocess it into a format from which useful condition indicators for a specific asset can be extracted, inject condition indicators into an ML model, and finally deploy the algorithm, initially to pilot assets to test the algorithm.
Sixth, corrective automated tasks and instructions to technicians are formulated to respond to alerts about possible component failures.
A commonly used framework for developing a predictive maintenance program is the reliability-centered maintenance (RCM) model. Not all equipment may be cost-effectively maintained using preventive maintenance techniques and not all equipment needs predictive maintenance. RCM helps organizations to analyze the main causes of potential equipment failures and design more efficient maintenance programs for different types of assets and different requirements.
The objectives of an RCM analysis are to identify failure modes, prioritize failure modes in terms of risk and cost, choose the best processes to mitigate failure, and preserve the desired functionality of a system.
RCM is not the same as predictive maintenance. Using RCM, organizations can analyze the benefit of different types of maintenance techniques, like predictive, preventive, or reactive maintenance for specific assets under different conditions, that is, for different failure modes.
For example, a particular piece of equipment might be prone to wear and tear on cabling and require predictive maintenance. Another piece of equipment might rely extensively on a reliable connection to the cloud and have alternative connection options in place to mitigate unexpected system downtime. Yet another piece of equipment might need regular, preventive maintenance to comply with strict safety regulations.
RCM was developed in the commercial aviation industry to improve the reliability and safety of equipment. It is based on the Society of Automotive Engineers (SAE) JA1011 standard, published in 1978 by the US Department of Defense. SAE JA1011 is the most widely used standard to develop predictive maintenance systems.
- Reduces the number of production hours lost due to maintenance
- Reduces unnecessary maintenance because equipment can be shut down just before a predicted failure
- Increases equipment lifespan and reduces resource wastage
- Reduces downtime due to equipment failure
- Helps system analysts to refine operations, for example by hardening redundancy processes
- Helps purchasing departments to make informed decisions about the most cost-effective hardware for specific tasks
- Supports compliance with safety regulations
- Gives asset managers, plant managers, and system designers a deeper understanding of the real-time performance of an asset
- Improves product quality by streamlining the production process
- Requires more capital investment upfront, for example costs for implementing a predictive maintenance program
- Requires specialized knowledge to analyze condition-monitoring data, create predictive algorithms, and integrate local systems with CMMSs
- Reduces the occurrence of failures but simultaneously results in less failure data available to feed back into the system for reliable performance analysis
- Can be time consuming to evaluate multiple assets
- Data can be misinterpreted and might not take all possible variables, like equipment age or environmental conditions, into account
- Requires maintenance team managers to make decisions and create work orders, increasing the possibility of human error
- Technologies like infrared, oil, and vibration analyzers are costly for small businesses
Designing and implementing a predictive maintenance solution can be time consuming, labor intensive and costly. If you are looking for a monitoring tool that supports you with your predictive maintenance program, contact the monitoring experts at Paessler to chat about your requirements.
PRTG can monitor almost any object that has an IP address. Read a case study about how PRTG products monitor the entire infrastructure for a water management plant here.