Most people associate predictive maintenance with hardware (sensors and networks) and software. Although the technology is an important part of the implementation, the process must start with clear objectives, risk analysis, and the application of good old fashioned common sense. In other words, predictive maintenance must be planned thoroughly and configured correctly by people before the technology takes over.
In this article we summarise the key takeaways from an interview with Johan Cannaerts, an engineer with specialized industrial service provider ERIKS, focusing on the best practices for the planning and preparation, as well as the role of IT, data, and analysis in predictive maintenance.
Let’s start by looking at condition-based maintenance. Cannaerts describes measuring and taking inventory of what is going on with your asset. For instance, if you are a manufacturer with several CNC machines, you will monitor and assess the condition of the machine tools, the motors, the electrical components, the computer network, etc.
Under a condition-based maintenance program, the maintenance is carried out when the condition of the asset demands it i.e., just before failure. To do this effectively you need to be able to predict when the failure will occur with a high degree of accuracy. The accuracy of predictions and the timeframe – minutes, hours, days, months, or years in advance – depends on the type of equipment.
As an expert in the application of IT and service solutions in an industrial maintenance setting, Cannaerts’ role at ERIKS is to drive technical services that help to optimise industrial operations and production.
Why predictive maintenance?
The main reason to adopt a predictive maintenance strategy is to reduce risk and damage in a factory setting. Risks include safety of people, environmental damage, and impaired functionality of equipment.
Cannaerts points out that customers such as factories and manufacturers do not have unlimited budgets and manpower, so they need to map out the greatest risks and pay more attention to those. Some failures are critical, others do not pose much danger when they occur. The cost of downtime is also a factor, and you need to estimate the risk and financial implications of a standstill.
“The scale and cost of repairs should be taken into consideration too. For instance, if you repair a machine before it fails you may only need to replace one bearing or gear wheel. If you leave it too long and the machine fails, you could be faced with replacing a whole gearbox or the entire machine. Also, spare parts that are engineered to order can take months to arrive, slowing down production. To alleviate this, keeping a supply of spare parts could form part of your maintenance strategy.”
The preliminary process
“We need to distinguish between analysing the data itself and dealing with the information that comes from it,” says Cannaerts. Therefore, the preliminary process should start with a risk analysis, but before investing in technology, you need to decide which equipment is most at risk and which needs the closest monitoring. “The best risk analysis method to use is a FMECA (Failure Mode Effect & Criticality Analysis) or similar methodology. Maintenance technicians and reliability engineers are able to perform this kind of analysis, or a third-party company can be brought in to carry it out and ensure compliance.”
In some industries, government regulation dictates that risks must be analysed and controlled. In this case, the company must demonstrate that analysis has been done and control measures are in place to detect and rectify defects or failures. If the failure mode is known, it can be detected and investigated with a sensor. Approximately three quarters of rotating equipment failures can be detected by vibration measurements. Thermography and ultrasonic measurements are also frequently used detection methods. With ultrasound, leakage of steam traps or cavitation can be detected, for example.
“The trend is more important than a current value.” points out Cannaerts, which means that monitoring should be carried out and recorded frequently, not just periodically. “Determining the correct sampling frequency requires specialist expertise. Vibration data produces beautiful 3D graphs, but what do they say? IT systems and software are useful to collect and store the data, but engineers are needed to interpret it. The data needs to be related to physical phenomena.
Therefore, algorithms should incorporate engineering expertise. For example, you can correlate a certain frequency with an imbalance in the machine. An IT specialist can write an algorithm to catch that logic, with input from an engineer. If that is the only anomaly, you know you need to balance the machine. For cases with multiple errors, such as electrical breakdown, imbalance and other forms of failure, you definitely need engineering expertise to interpret the data.
A company can invest in building up the required expertise itself, but the question is whether such a specialist has a full day’s work. Alternatively, they can choose to outsource the expertise to specialised companies. Such companies almost always work with cloud solutions, which issue a weekly or monthly report and recommendations or offer a dashboard which gives information on the state of the machines.
For example, a large oil company has a large number of refineries, chemical plants and drilling platforms. For them, it pays to have that expertise in-house. ERIKS also has enough scale to recruit experts, but one problem is that there is a chronic shortage of specialists. Software can help to fill this gap to an extent, but you still need people that can critically test the algorithms, set the correct sampling frequency, analyse the data correctly, and know what to do if certain parts of the machine fail.”
Starting point of predictive maintenance
“You always have to ask yourself: if this part or that machine breaks down, what happens then?” says Cannaerts. “If a customer has three parallel production lines and all three end up at one machine, then clearly that machine is critical, and you must monitor it. You can then carry out maintenance when it is convenient. If you realise that certain activities recur every month, you can immediately include that machine. Do you really need to apply condition-based monitoring in that case? You need to find a balance.”
The same applies to components. “Do you have one machine with a specific electric motor, or several factories each with a whole fleet of machines with hundreds of the same motors? Then it may be interesting to invest in condition monitoring and to analyse whether those motors drive certain functions well. Perhaps you want to find out whether certain brands of motors perform better or worse than others, and why.
Most of the money for predictive maintenance goes into hardware and software. The great thing about cloud solutions is that you can pay per month, instead of investing in expensive software and spending a long time installing it. With a cloud solution, you pay for its use and if you find you don’t need it after a while, you can cancel your subscription. This reduces the cost of the software considerably.
Mounting, installing and configuring sensors and other hardware can cost a pretty penny, especially in ATEX environments. Connectivity on site can also be tricky. You can use 4G and 5G, or LoRa or ZigBee and protocols like that, but you must test in advance whether you have the required connectivity. Also, the hardware must be suitable for that environment, whether wired or wireless.
It costs a company a lot more when critical systems fail, resulting in lost production. The mindset towards condition monitoring has certainly changed, because factories also have fewer specialist staff on the floor to carry out preventive maintenance. COVID has also played a role and we are seeing growing market demand for predictive maintenance.”
Who carries out predictive maintenance?
A common question is – which people in a company manage and work with a predictive-maintenance system?
A distinction needs to be made between analysing the data itself and dealing with the information that comes from it. The vibration analyst may detect an imbalance in a shaft, for example, but the maintenance department need to do something about it. That can lead to a degree of tension between maintenance and shop floor staff, especially if the roles are not clearly defined.
In many cases, condition monitoring can be used to inform the production department, who can analyse what the consequences may be. For instance, they may decide to continue production at full speed and if a piece of equipment or machine breaks down, a whole batch may be destroyed. Someone will have to decide whether to stop the production or continue.
Suppliers of hardware and software prefer to sell as many sensors, networks, edge computers and software licences as possible. Cannaerts shared his view on this, “Our specialist Hanno Spoelstra already provided input for a study by PWC and Mainnovation in the Benelux in 2017, for example. The majority of data retrieval for predictive maintenance turned out to be done by people in the field with a measuring device. Recently, an online poll was conducted during a webinar to update that research. It also showed that visual inspections in combination with mobile measuring devices are still the most applied form of collecting data for predictive maintenance.”
“I have seen examples of this myself. For example, ERIKS does surveys of steam traps, usually once a year. We looked for sensors that could detect whether the traps were leaking. We found one but it costs more than a hundred times as much as the steam trap itself. So, machines that do not fail easily and are less critical should not be equipped with sensors. Technological developments will continue though as we will have fewer and fewer people in factories carrying out all kinds of inspections. But, predictive maintenance will also always be: the worker in the field taking measurements with a thermography or ultrasonic device. Some things still rely on engineers using their own senses: look, touch, smell.”
The role of ERIKS
ERIKS has set up a separate team for asset management. The reason for this is that they have a lot of knowledge about the solutions and products they sell and are well placed to analyse the condition of machinery and advise on how to prolong service life.
ERIKS realised that there is a growing need for this expertise that businesses do not have in-house. Therefore ERIKS provides a cost-effective expertise service as an add-on to the products themselves.
Cannaerts explains, “Steam traps are a good example. They are found in many installations, but customers sometimes don’t even know there are steam traps in their system, let alone where they are located. You need an expert to listen and make statements about installation, leakage and energy loss. The latter is not just about saving money, but also about reducing CO2 emissions.”
“It is mainly about parts for machines on a functional level. A pump always contains a bearing and the vibrations it emits say something about the whole installation. Cavitation, a bent shaft, dry running and electrical breakdown can all be seen in the vibration pattern.”
He continues, “Hoses, for example, must be inspected annually for compliance reasons. This is also a form of monitoring: you carry out a pressure test, just as you do when a new hose is delivered. In combination with a visual inspection for damage such as cracks or dehydration, you can assess whether the hose is still good or needs to be repaired or replaced. We at ERIKS understand that people remain an important part of predictive maintenance and it doesn’t all have to be digital straight away. “
“Humans remain an important part of predictive maintenance.”
ERIKS provides a contract monitoring service, as predictive maintenance needs to be carried out over a number of years.
“Only when you see a trend, you can determine when parts or machines need maintenance or replacement, which requires a longer-term contract. For example, ERIKS has recently acquired a large project for all the valves at a terminal. When they deliver valves, they are assembled, tested and delivered with a test certificate, which immediately provides a baseline measurement on which to base the predictive maintenance.”
On October 27, Johan Cannaerts will kick off the Asset Performance 4.0 conference with a keynote speech on predictive maintenance.