The philosophy and your benefits
What would you do if you could look into the future?
If your main concern is to keep your production running, you might answer:
“Predict any upcoming machine breakdown so I can avert it.”
This is the core philosophy of Predictive Maintenance. It is also a central concept of Industry 4.0. Upcoming machine issues are forecasted by looking at their sensor data.
The goal is to schedule proactive maintenance just in time to avert downtime. The learnings can flow into a better prediction and prevention of future issues.
NASA made the first effort in predictive maintenance. Their goal was to avoid faults of the Space Shuttle’s main engine. The concept has been gaining traction in other industries ever since. Early adopters on the forefront apply the concept to wind turbines and jet engines.
First adoption comes from industries which value high machine uptime and little maintenance. Data driven companies (such as IBM, GE, SAP) carry the concept to other industries. Examples include the automotive and semiconductor manufacturing, which run mission critical machinery.
Predictive maintenance differs from traditional concepts like reactive or preventive maintenance. It reduces waste in time and equipment while minimizing risk of downtime. For this, it is necessary to collect, store and analyze data sets.
Sensor data is the foundation for all diagnostic information. Insights are then generated in the machine itself or by the equipment manufacturer.
Why Predict?
The Core Applications
With predictive maintenance, not only scheduling of maintenance tasks becomes more precise. Manufacturing companies and equipment providers can also offer a new range of applications.
An example is the monetization of optimization services. A better understanding of asset performance improves forecasting as well as optimization.
1) Monitor
The stream of sensor data gives insight into the health and load of a machine. With key data streams available, it becomes possible to watch for anomalies.
These can be data trends, spikes or breaks. Knowing how the data relates to real asset behavior can help to visualize the plant status at all times.
2) Forecast
Historical insights added to most recent data fuel the extrapolation of future events. The goal is a forecast of:
Which assets will most likely fail
The time at which they need servicing
The reason for the failure
Early failure prediction enables scheduling of counter actions. This can be carried out at a time not causing production disruptions.
With a good maintenance forecast, the coordination of field service becomes most efficient. The deployed maintenance personnel is on time and has all relevant information. So, all actions are clear and the needed spare parts are on hand.
The planning of tasks and spare part usage helps to optimize the spare part inventory as well. This leads to reduced cost of stockouts and overstocking.
3) Optimize
Sustainable production relies on a reduction of waste during asset life time.
For this, the transparency and optimization of the ongoing process is key.
Failure analysis helps to understand underlying causes and avoid them in future. The analysis method is also known as “root cause analysis”.
The collected data and information about key performance drivers allow for process optimization. So, an increase of asset utilization, product quality and process efficiency becomes possible.
An increase of production performance also benefits the company through higher customer satisfaction.
How does it work?
On a high level, predictive maintenance relies on three layers. They need to be in place for reliable predictions:
The accumulation of relevant data, interpretation with guided learning, and finally predicting events.
The Road to Proactive Maintenance
1) Run to fail
From own experience, this is still a common practice in many production areas.
One reason being the lack of skilled maintenance personnel. Another reason is the common credo “don’t fix it if it ain’t broke”.
The troubleshooting and problem analysis only follows when errors or incidents happen. Reactive maintenance causes high unexpected downtime.
Additionally, a long waiting time until the needed spare part is at hand.
2) Preventive
Preventive maintenance concepts make use of a fixed schedule maintenance planning. Time between maintenance jobs derives from experience values or equipment manufacturer guidelines.
For slightly better results, the maintenance jobs can follow a cycle based schedule. This would notify the maintenance crew after a certain number of production cycles. The inspection results and observations help to adjust fixed maintenance intervals during operation.
3) Condition based
“Condition based maintenance” uses the asset condition to build the maintenance schedule.
The asset health indicator and load factor are important indicators. They are used in combination with key wear drivers to adjust the target counters.
For example:
A machines “health” lies at 67% with a threshold of 60%.
The “health” factors in:
Time since last maintenance and inspection
Production cycles run per day
Oil temperature readings
The oil change is set to be due 3 days from now, as 2.5% health are “lost” per average production day = “load” of 2.5% per day.
4) Predictive and proactive
Proactive, data driven maintenance predicts the correct maintenance action and timing. The forecast uses the incoming stream of process data.
The next step is a combination with historic learnings and business data. This allows an interpretation of asset condition using advanced analytics.
A main concern with data based maintenance is the unpredictable influence of human actions. In some cases, the concept of analytics takes in user feedback. This allows to fine-tune the algorithms and adjust for inaccuracies.
If high quality data is available, data science methods can be used to find correlations.
In some cases not even experts are aware of the uncovered findings. With the integration of more data sources and big data analytics even more insights can be generated.
The Difference between Preventive and Predictive
Three factors characterize the difference between preventive and predictive maintenance:
1) How is maintenance triggered
Preventive Maintenance: Triggered by time, events, cycles, or other readings. It factors in the age of the equipment, and recommended service intervals. But the real wear of the machine is neither known nor used. This can lead to unnecessary maintenance and use of spare parts.
Predictive Maintenance: Determines the equipment condition to predict machine failures before they occur. This leaves the team enough time to schedule and carry out the needed service tasks.
2) How is maintenance conducted
Preventive Maintenance: The trigger for a maintenance job leads to a complete maintenance job done. This means, parts are sometimes exchanged before they show real signs of wear.
Predictive Maintenance: Advanced techniques like visual, acoustic or vibration are used. They track and analyze the asset condition. So it becomes clear which part exactly needs servicing without prolonged troubleshooting.
3) Costs and Savings
Generalization of costs and savings are not easy. One source calculates the relative cost of preventive ~30% higher than predictive maintenance.
Preventive Maintenance: Unnecessary maintenance tasks and premature exchange of spare parts. An extra risk lies in the technician causing damage during one of the frequent repairs.
Predictive Maintenance: Parts are only replaced when the explicit need arisis. No unnecessary work gets done and the parts are best employed. It can be costly to install, but some reports in the oil and gas industry list an ROI of up to 10x.
Technology
1) Network
Predictive maintenance relies on a steady data stream, supported by an appropriate network.
Local applications may rely on standard TCP/IP via Ethernet for stability reasons. Long range applications may use wireless sensor networks, minimizing the cost of wiring. (i.e. LoRaWAN, LPWAN, cellular)
2) Big data infrastructure
The data for predictive maintenance consists of varied formats and vast data sets. The use of intelligent algorithms and storage of insights requires the right infrastructure.
Big data applications face similar challenges. They rely on a high storage and performance data base system.
3) Sensors
Sensoric capabilities are needed to automated machine inspection. They use non-destructive technology to sense indications for maloperation or upcoming failures.
Additionally, process performance data is gathered by other devices in the manufacturing chain.
The use of vibration sensors and analysis is common in high-speed rotating equipment. This is a good indicator to sense increasing wear of bearings before their breakdown. But the operator still needs practical knowledge of vibration analysis.
An asset’s outward appearance can be a show of wear, if cracks or corrosion become visible. Cameras combined with good ambient lighting can fulfill the tasks, also remotely. Visual monitoring can also deliver infrared sensing capabilities. This can spot mechanical or electrical issues. Infrared offers a wide application range at an attractive price.
Acoustical analysis with sonic, or more common, ultrasonic frequencies. These sensors can help to “hear” stresses in rotating components. A technique for an early failure prediction compared to other sensor technologies. The friction waves can be identified as ultrasonic signatures. If these patterns change, they can indicate a changing state. Which means damages, bad lubrication or the start of wear.
Chemical composition sensing analyzes the machine oil constituents using sophisticated sensors. It scans for the condition of the lubricant and for wear particles. But to reach a sophisticated wear analysis based on particle concentration takes years to master.
Spectral analysis of electrical current and voltage signals shows deviations from the norm. Which in turn can point towards electrical and mechanical issues.
Examples
A common application is monitoring in wind turbines and aircraft engines. Their high cost for downtime and hard to access installations benefit from precise maintenance.
Typical data sources are rotation monitoring, (ultra)sound, vibrations, as well as temperatures. This allows to prevent worn out bearings in wind turbines, one of the major tasks.
Another everyday example is the prediction of maintenance in automotive. The motor, suspension, tires and other parts are continuously monitored to forecast the ideal service date. This avoids downtime and expensive repairs.
Also spare parts can be exchanged just in time during an already scheduled visit to the shop.
Connected cars can also send valuable information about spare part needs and upcoming service dates to the original equipment manufacturer.
Final Thoughts
After only just touching the surface of what predictive maintenance is all about. Your considerations as equipment manufacturer or production company should also contain:
Which data can be extracted and accumulated from the products in use?How to upgrade them with sensor and communication technology to make them ready for condition based maintenance? And are we able to capture the wear indicators in a reliable way or is the maintenance of assets influenced by a combination of complex factors? A complicated and complex influence on maintenance calls for a big data basis to deliver useful algorithms.
And of course, “Is it worth the effort?” The answer to that depends on the cost of downtime and of current inefficiencies in the process. Yet, you don’t need to go all-in with predictive maintenance. Start an initiative to move from “run-to-fail” to preventive maintenance. After that, recap your investment and discount future savings and revenues . This will give you a foundation upon which to base further investment decisions.
Can you form a business model from your predictive maintenance capabilities? Equipment providers leverage the data driven insights to offer new, value-adding services. For example guaranteeing a certain level of availability. These services are only possible with a capable and flexible workforce.
Daniel Sontag connects the bots: As Industry 4 lead and manager for connected products, he does what he loves — tying business to tech, and theory to practice.
Hi, great you enjoyed the article! Feel free to like or leave a short response below, thanks.
Stay tuned: JumpStart-PM, The Industry 4.0 Blog and on LinkedIn