Are you working for an industrial company that is discarding most of its data? Are you wondering whether that data is valuable? We can help you decide.
Predictive maintenance techniques and software have evolved over the years. Companies do not need to rely on techniques where they import data to spreadsheets and uncover insights manually. With the increasing usage of AI and machine learning algorithms in predictive maintenance tools, now, businesses can accurately predict maintenance tasks thanks to big data.
Advanced equipment such as Internet of things (IoT) and Industrial Internet of things (IIOT) and robots are generating more data than ever before. However, factories’ analytics capabilities have not caught up to this. According to Fero Labs, manufacturing companies discard 98% of all the data they can collect because they do not have the operational analytics capabilities to integrate that data into their operations.
And are they right to discard that data? No, according to McKinsey&Company. McKinsey predicts $1-4T value creation p.a. in 2025 due to IoT in the factory setting globally. A large portion of that value creation is due to industrial analytics and predictive maintenance. Naturally, we would expect hype around these impressive values and according to GE, even back in 2014, big data analytics was one of top 3 priorities for >84% of CxOs in factory settings.
Predictive maintenance (PdM) and industry 4.0 companies step in to fill the gap between data and insights for industrial companies. Predictive maintenance software allows companies to store and analyze critical outputs of their machinery. One of the key things to do with that data is to improve maintenance and input parameters of their machinery.
What is predictive maintenance?
Predictive maintenance (PdM) is conducting maintenance to prevent predicted problems rather than conducting maintenance on a fixed schedule or when an issue arises.
You may have heard predictive maintenance but may not be sure about how it is different from your current practices. Let’s try to quickly explain other maintenance practices and help you understand why predictive maintenance is a better option.
- Planned maintenance, preventative maintenance or scheduled maintenance: Maintenance schedule is fixed, maintenance is completed at regular intervals. Maintenance activities are fixed, complete control and maintenance of all machine components are conducted over time. Not all machine components may be checked with the same frequency though. So maybe component A is checked monthly, component B is checked annually. However, over time, all fault-prone pieces of equipment are checked.
- Condition-based maintenance: Maintenance is conducted as issues arise.
Predictive maintenance is better than these other approaches because it allows the organization to prevent problems without incurring the cost of unnecessary frequent maintenance. So over the lifetime of a manufacturing facility, some components may never be checked if they are not predicted to cause problems.
Executives rarely ask “What is predictive maintenance?” because it is so intuitive. If we knew when malfunctions would take place, we would do maintenance at the first stop before the malfunction. That’s the exact premise of predictive maintenance and industrial sensors, this is becoming possible.
Though predictive maintenance seems intuitive, it may not be worth setting up predictive maintenance if its benefits will be negligible. In short, if up-time is not critical, maintenance costs are not significant and advanced analytics is unlikely to bring significant value to business, predictive maintenance may not be worth it. For a more detailed discussion, please see our guide which explains the maintenance approach to choose depending on your goals.
Why is it important?
Asset management is critical in industries like manufacturing where advanced machinery is expensive and depreciation is a major cost. Major cost savings are possible with predictive maintenance. Essentially, predictive maintenance is the new lever in improving asset management. While six sigma and lean management were older methods for driving efficiency, after having been used for more than half a decade, they have limited returns for today’s companies.
According to PwC report, predictive maintenance in manufacturing could
- improve uptime by 9%
- reduce costs by 12%
- reduce safety, health, environmental & quality risks by 14%
- extend the lifetime of aging assets by 20%
How does it work?
While the general idea of predictive maintenance is intuitive, predictive maintenance systems rely on a myriad of sensor data for machine condition monitoring. Some of those sensors measure:
- Rotation speeds
- Chemical properties oil
Depending on the machine, above or below normal values in these sensors can signal future issues and creates work orders to perform maintenance. For example:
- Increased temperatures can lead to components melting or burning and depending on the equipment may need to be remedied before causing significant damage
- Vibration analysis could offer insights into possible breakdowns as increased vibrations can be signs of component failures.
- Oil analysis involves analyzing a lubricant’s properties to estimate how the machine is depreciating. Depreciation rate is estimated by measuring the amount of suspended contaminants, wear debris and so on in the lubricant. While 50 years ago, this was the domain of tribologists, machine wear experts, now machine oil is extracted automatically by IoT and analyzed either by IoT or in lab setting revealing detailed wear patterns.
Real-time monitoring of these variables allow immediate interventions to solve issues before they arise. Additionally, time series analysis can highlight abnormal deviations. While these deviations may seem acceptable when examined in isolation, time series analysis can identify abnormal patterns and predict future problems.
Let’s look at how those data, for example increased vibrations, get translated into maintenance actions.
How did industrial analytics & maintenance software evolve?
Maintenance software evolved with data processing capabilities of the day. Starting with productivity tools like excel, maintenance now relies on advanced analytics.
Most industrial companies today are still in this stage. Lean or Six Sigma teams get data from relevant teams, crunch the numbers in excel, SPSS or MATLAB, create insights, communicate their results, get buy-in and help roll-out of their proposed changes and record results to see some improvement. Finally, in a closing meeting, results are shared and a new way of working becomes a part of the company’s processes.
This approach is labor intensive, not scalable and depends heavily on the team involved. However, most companies still operate in this fashion, consultants still sell Lean and Six Sigma projects. Management, like science, advances one funeral at a time, especially in oligopolies where short term survival of companies is guaranteed. However, long term decline of such manual approaches is quite evident:
Funny side note, tried to have the same graph for keyword lean. Lean is surging in popularity! Increase in people’s desire to get fit dwarfed any trends due to Lean management practice.
While manual methods provided the possibility of constant improvement, Computerized Maintenance Management Systems (CMMS) allowed dispatching technicians, tracking their progress and control overall maintenance effort. CMMS provides initial data of the maintenance cycle and generates alerts and work orders whenever it identifies a machine is operating outside pre-defined conditions. While CMMS is clearly more efficient than spreadsheets, historically CMMS systems still did not provide significant analytics capabilities. This has changed with modern CMMS companies such as eMaint and Limble offering strong analytics capabilities
Virtual physical twin
Engineers with detailed knowledge of an industrial machine’s working principles, build a mathematical model of that specific industrial machine. These models are fed real-time data to produce insights. Because they need to be custom-built for each machine, they are not scalable across the factory floor. Additionally, they can only be produced with sufficient accuracy by the teams that built the machine, reducing competition for this service. Therefore, they are practical only for organizations without more generic analytics capabilities that want to enable predictive maintenance on a few critical systems.
Virtual statistical twin
Data scientists leverage historical sensor data to build a statistical model of that specific industrial machine. While machine-specific knowledge is not required to build a virtual statistical twin, this approach is also not scalable as it is machine-specific and requires expensive data science talent to build and maintain twins.
Assisted learning systems
Machine learning systems complete predictive analytics and highlight patterns in sensor data. Technicians match auto-generated insights with events on the factory floor and choose patterns that need to be identified. Machine learning models are not machine-specific and can be deployed across the factory. In this model, there is no need for model maintenance work so engineers and data scientists do not need to be employed. Most currently sold predictive maintenance solutions fall under this category.
Future of industrial analytics and actually any operational system. A fully autonomous system making real-time decisions, monitored only when anomalies take place. Infrequent audits could be conducted by humans to ensure safe&sound operation and understand how autonomous system could be further improved. This is very rare in practice yet in 20 years, any other approach will probably seem quite arcane.
What are its benefits?
One of the most comprehensive studies on potential of industrial analytics was conducted by McKinsey in 2015 and we used their estimates in showing how much improvement is possible. Their study was based on client studies so we believe it can give you a good idea of the potential:
- 50% reduction in downtime due to equipment failures: Asset failures are costly and stressful. An hour of downtime can impact millions in revenues for a $100 M+ company. Since issues can be predicted in advanced, downtime can be minimized. Increasing up-time is a significant challenge for manufacturing or logistics companies with machines that depend on one another. An example is downtime in cranes. As Predikto CEO Mario Montag mentioned, ports experience 800 to a thousand hours a year in downtime due to crane malfunctions which is extremely costly for port operators.
- 3-5% increased machine useful life: Since predictive maintenance reduces machine breakdowns and ensures operation in optimum settings, it can improve machine/robot useful life.
- Reduced environmental impact: As machines remain useful for longer periods and as their efficiency increase with advanced analytics, companies will waste less natural resources. Predictive maintenance is one of the few initiatives that both help companies bottom line and their corporate social responsibility goals.
- 10-40% reduction in maintenance costs: Since planned maintenance is based on a schedule, there will be cases when maintenance tasks will be performed when they are not needed. Predictive maintenance can prevent such inefficiencies. Furthermore, predictive maintenance systems inform technicians about the changes they need to do to the system based on symptoms. For example, let’s assume that sensors show increased vibration is observed in a machine. If there’s a strong correlation between malfunction of a specific part and increased vibration then technicians can focus first on the possibly malfunctioning part, completing only necessary maintenance activities, saving time.
- 10-25% reduction in worker injuries: Leveraging sensor data with analytic systems will help industries find new ways to avoid injuries. Reduced breakdowns and accident avoidance systems that can alert or even halt equipment when there is a danger to a worker, can dramatically improve factory conditions and minimize worker injuries.
- 10-20% reduced waste: Sub-optimal operation that is not detected, can result in wasteful production. Raw material, energy, labor costs and machine time get wasted in such instances. Predictive maintenance systems can uncover issues that can result in waste before they arise.
- Advanced analytics: Setting up predictive maintenance involves collecting sensor data from diverse machinery. Once that data starts to be automatically collected, analysts have a trove of information ready for analysis. This data can be used to identify parameter and process optimization opportunities.
- Improved product quality and increased customer satisfaction: Detailed sensor data and ability to observe results of interventions create a virtuous cycle of experimentation and learning. As teams adjust machine parameters and improve results, they uncover means to improve quality.
- Increased employee morale: Downtimes, operation with sub-optimal parameters not only impact output but also impact employee morale. It is stressful to rush to solve problems when they arise. Predictive maintenance minimizes such instances.
- Improved performance over time: Predictive maintenance systems are learning systems. They implicitly create a knowledge base of issues and understand their root causes based on feedback from technicians or sensors on the shop floor.
Which industries can benefit from it?
Predictive maintenance can be applied to all industries where machines produce significant amounts of data and require maintenance or fine-tuning of their parameters. Both discrete industries like consumer packaged goods (CPG), automotive, electronics, textiles, aerospace and process industries like food and beverage, chemicals, oil&gas, pharma can be transformed with predictive maintenance. A significant share of vendors are actually industry agnostic and serve most industries as their work relies on data interpretation and can be abstracted from the specifics of machinery in the factory floor.
An overview of industries where predictive maintenance applications are already gaining traction:
- Automotive: Automotive companies operate some of the largest robot parks in the world. With the aim to reduce inventory costs, automotive companies developed Just-In-Time manufacturing methodology since the 1960s and 1970s. As a result, they have tightly integrated supply chains. Though tight supply chain integration allows reduced inventory, any reduction in manufacturing efficiency results in significant disruption to the supply chain. It is no surprise that automotive companies stand to gain significantly from a technology that reduces downtime.
- Airlines: Airlines are no stranger to closely monitoring sensor data from planes. Today’s analytical capabilities allow them to ensure the safety of passengers by analyzing more data.
- High tech manufacturing: Operating complex equipment at optimal parameters is the key challenge to improve efficiency for high tech manufacturers like semiconductor manufacturers. Predictive maintenance systems allow them to operate at a level closer to optimal parameters.
- Transportation: Though airlines lead the pack in terms of complexity of their equipment, other means of transportation like trains also involve complex machinery that can benefit from predictive maintenance.
- Oil&gas: Despite the rise in green energy, oil&gas is still one of the largest industries. Both extraction and refining involves expensive equipment that can cause health and environmental hazards in case of failure. For example, Deepwater Horizon oil spill in 2010 which led to 11 dead and ~5 M barrels of oil spilled, has been one of the worst disasters in the last decade. Stakes are high to prevent such disasters with better analytics and maintenance.
- Ports: Exposed to harsh conditions, port equipments’ conditions deteriorate quickly. For example, cranes are critical components but they are prone to failure. Crane downtime means more waiting time for ships and less throughput for ports. Reducing downtime improves service quality and reduces waste for ports.
How to implement a predictive maintenance program?
Here’s how you implement a predictive maintenance program: There are 2 common ways to roll-out a predictive maintenance solution.
Building an in-house solution leveraging open source libraries like Python. Depending on your company’s skill-set, this could be attractive. If you have engineers willing to spend their time off work to play with sensor data and predict failures, you could end up with a rudimentary predictive maintenance solution for free.
However, in most cases, engineers do not have time for such large side projects and you would need to hire a data scientist. You could even hire a specialized predictive maintenance engineer but this path will probably end up being expensive and slow.
The alternative is buying a predictive maintenance solution that has built-in integrations with your advanced machinery allowing you to get up-to-speed quickly and paying as you reap rewards.
Once your software solution is in place, a bit of organizational work is needed to reap the full rewards. Firstly, all engineering teams should be notified of the advanced analytics capabilities you needed to build for predictive maintenance. For example, difficult-to-measure improvement programs like Six Sigma or Lean need to be monitored with your firm’s new analytics capabilities. Secondly, maintenance teams need to learn to use the full analytical capabilities of the solution. It is not efficient if they continue to do a complete checkup on robots when a slight adjustment on a specific component is needed.
If you are planning on working with a vendor, we outlined the industry landscape for you:
Understand vendor landscape in 2 minutes
To choose a suitable vendor for your business, you must understand the vendor landscape and compare vendors. There are mainly 4 types of predictive maintenance vendors:
Companies like IBM, SAP, SAS are in this category. Their primary advantages are existing business relationships and software expertise. They are not focused on specific industries and their domain expertise is limited compared to other predictive maintenance vendors.
Industrial automation leaders
Companies like GE and Siemens are among few companies that supply industry with automation equipment including advanced robots. They have the deepest customer relationships and domain expertise as they build the tools used by the industry. However, software, especially end user oriented software has not been the strongest point of these companies. To compensate for this, they are building large teams of data scientists and strongly promoting their products.
GE is also looking to leverage other companies in its industrial analytics and predictive maintenance efforts. Its Predix platform allows other companies to build analytics solutions on top of data collected on the Predix platform. If the platform can become ubiquitous, it can become the App Store of industrial analytics.
Industrial analytics & predictive maintenance leaders
Established in early 2010s, companies like Augury, Falkonry, Predikto, Sight Machine are focused exclusively on industrial analytics. Their solutions are quick-to-deploy and integrate with existing industrial machinery. They focus on providing advanced analytics in an easy to use manner. These are much smaller companies compared to the companies above like IBM and have limited sales teams so they are easy to miss if you are not focused in your research.
This is still a very hot field. While trillions of value creation is predicted, actual impact lags significantly behind the predicted numbers. Therefore it is not surprising that new companies are entering the foray with support to the tune of millions from VCs. They aim to differentiate themselves with more advanced machine learning capabilities, focus on specific industries or more intuitive interfaces. With researchers uncovering new machine learning techniques, we expect more startups to be founded by industry leaders partnering with researchers to commercialize new technologies.
How is it priced?
There are several possible approaches to pricing predictive maintenance services and products. Vendors follow different approaches:
- Value based pricing: Ensures maximum value for vendor while aligning incentives of vendor and customer. It is not popular since it is difficult to objectively and accurately estimate or measure value. It is more efficient to agree on a simpler pricing model.
- Fixed + variable pricing: Most commonly used model as it reflects vendors’ cost structure. It is a significant effort to set up a predictive maintenance system because it involves pulling data from diverse robots and machines. However, adding additional robots or additional analytical packages can be relatively cheaper. Fixed portion of the price ensures that vendors can profitably serve customers who are using the service on a limited set of machines. Variable portion of the price ensures that new machines can be added to predictive maintenance at a lower cost per machine.
Though we believe that predictive maintenance is one of the most important AI use cases, especially for manufacturing companies, there are still other AI use cases in operations that can transform your business. You can see our section on AI in operations.
Evaluating vendors and making the right vendor assessment can be time-consuming. If you are short on time and want to work with experts who can suggest you the most suitable vendors for free, just let us know:
Predictive maintenance of critical machinery can have substantial benefits. It maybe worth building a custom solution, specific to your companies machines. We can help you find the right partners for that:
How can we do better?
Your feedback is valuable. We will do our best to improve our work based on it.