6 Tools for a Successful Predictive Maintenance Program
Now that the Big Data hype has slightly eased, data practitioners are realizing that while we can, or do, collect large volumes of data, it is more important to be tactical with the data we are using. But a seemingly new contender for the data spotlight has sprung up: IoT data.
But is IoT data and the attempt to adapt it for data driven decision-making all hot air? Or is there actual substance to it? We at Hitachi Solutions argue that the IoT takes one of the more tractable and immediately beneficial aspects of modern data capture, high velocity data, and helps us drive highly responsive and, when utilized effectively, pre-emptive data driven decision-making.
Predictive vs. Preventative Maintenance
It is important to distinguish between predicative and preventative maintenance. Unlike preventative maintenance which seeks to decrease the likelihood of a machine’s failure through the performance of regular maintenance, predictive maintenance relies on data to determine a machine’s likelihood of failure before that failure occurs. This allows manufacturers to move from a repair and replace model to a predict and fix maintenance model using predictive analysis — which relies on data, statistics, machine learning, artificial intelligence and modeling to make predictions about future outcomes.
The Rise of Predictive Maintenance Programs
In recent years, predictive maintenance has risen to the forefront of the IoT data industry for two key reasons:
- First, modern machinery often comes with embedded computer chips to control the machine and also take readings from these machines. Although this data hasn’t always been stored by the machine owners, it is available for them to start capturing.
- Second, with new machines, their embedded sensors, and new information technologies for inexpensive capture and storage, the cost (data, effort, and monetary) to implement predictive maintenance has and continues to be significantly reduced.
This means that in the next few years we will see more business problems, even those that are not machine maintenance specific, become restructured as predictive maintenance or predictive service questions. For example, the highly sensationalized story of smart fridges that are able to monitor our supply of food and ingredients could potentially lead to an automation of our consumption choices. This opens the door for predictability in our expenses —and gives us the ability to better control our diets through planned food expenditures over the impulse buys that grocery stores encourage today.
I use this example, specifically, to demonstrate that predictive maintenance or predictive service programs and solutions will eventually solve problems we never even thought about — once we learn to creatively recast those problems in a way that can be solved using IoT data and predictive analysis.
Today, however, the most common use cases for predictive maintenance are in the following industries:
- Power Generation
- And certain service industries such as HVAC, Elevator, and ATM monitoring and maintenance (because these industries have been collecting information from their machines for quite some time now).
Additionally, the newer the machinery and technologies used, the more sensors and IoT data that is collected.
What Are the Objectives of Predictive Maintenance?
For most cases today, the objectives of predictive maintenance programs can be boiled down to one of two outcomes:
- Improving production efficiency.
- Improving maintenance efficiency.
Production efficiency can be improved by maximizing the time that machines are up and running through predictive maintenance, or it can be also be improved by predicting the number of goods that will pass or fail a quality inspection, based on the readings from a machine. Another, less intuitive but interesting example is how we could even predict the condition of a machine based on the defects it creates in its outputs; in this case, the predictive directionality is reversed.
We can make these predictions by taking the most recent data points from a device and running predictive algorithms against them. In the case of predictive maintenance, this enables us to customize our maintenance activities to each specific machine, or even for each specific component on a machine; a noteworthy departure from the traditional method of basing our maintenance off of schedules or usage thresholds, such as miles driven in a car.
For example, let’s imagine a renewable energy production facility such as a wind farm. We have enormously complex machines that are riddled with sensors and computer chips that not only control each wind turbine, but also send constant readings of the state of each component on each wind turbine.
In spite of this continuous stream of data, traditionally, turbines have been taken offline based on an annual maintenance plan, much like the regular checkup appointments that we take a car to. The turbine is taken offline and a maintenance crew ascends the turbine to do a full inspection of the mechanical and electrical systems, as well as a check of the physical integrity of the machine. These inspections are done, today, at routine intervals according to the specifications of the turbine manufacturer.
However, this one-size-fits-all approach to maintenance is inefficient, expensive, and is not sensitive to the conditions of the wind farm or the turbines. For example, if a turbine sits in a warm climate, the operational conditions as well as the stressors on the machine would be starkly different than the stressors on a turbine sitting in a northern location that experiences freezing winters and ice storms. If we take this a step further, there are even different stressors on each turbine depending on its location within a wind farm site (e.g., the lead turbine, exposed to the strongest gusts of wind on a ridge vs. a turbine near a reservoir, exposed to humid winds).
The rise of predictive maintenance is an attempt to be pointed and precise in the identification of risks and failures on our precious assets, and to enhance our ability to be responsive to the unique pressures on each machine. Consequently, each machine receives the maintenance that it needs, when it needs it, in order to keep it running and operating in peak condition for the longest time possible.
By adopting a predictive maintenance program, we give ourselves the opportunity to move away from a costly and inefficient maintenance cycle. We are not only able to increase the time our assets are operational by pre-emptively identifying problems before they occur and adversely affect the rest of the machine; we are also able to increase the efficiency of our maintenance activities, helping us to be more precise in the scheduling and deployment of our crews and the tasks they are performing on each deployment.
Predictive Maintenance Tools
Successful predictive maintenance programs all rely on a set of tools to make them function. While the tools may differ slightly from product to product, there are six primary tools and techniques that all successful predictive maintenance programs should have.
1. Small Early Pilot Programs
Before you invest in an entire predictive maintenance program, do pilots. Pilots will help you determine whether the value is there in your organization. Find a partner that has a SaaS service you can pilot your predictive maintenance program with. Start small and see what predictive maintenance can do for your organization and then iterate and grow from there. This is important because you want to ensure you establish the correct platform for your program. Once you have a platform in place that you know will work for your business, you can then expand from there.
2. A Technology Suite for Aggregating Data
Data is the engine of any predictive maintenance program. Therefore, there must be technology in place to collect, process, prepare and structure massive amounts of device data which will be stored in the organization’s ecosystem. This system must be able to understand what each piece of data represents so that it can be monitored as a part of the entire maintenance feedback landscape.
3. Algorithms to Monitor Patterns and Events in Real Time
Once there is data streaming in and being collected from your industrial equipment, data science and machine learning come into play. By monitoring patterns in real time and looking at historical data, the machines themselves can identify repeat scenarios which they can then create rules for moving forward. This process is an adaptive learning process, meaning that the machines learn over time. The more data and scenarios they collect and encounter, the more they learn.
For organizations just starting out with a predictive maintenance program this means that predictive capabilities will most likely not be available on day one, as the machines need to have a period of data collection and analysis before they can begin to predict future outcomes. But that doesn’t mean that you should sit around waiting for your machines. While your machines are busy building their knowledge, which is the critical and essential foundation for any predictive maintenance program, you can implement condition based monitoring. Condition based monitoring gives you the ability to immediately benefit from a predictive maintenance solution. With this strategy, you would set in place conditional algorithms, such as “if X falls outside of such and such parameters, then do Y.”
4. Effective Workflows
Once your machines have collected enough data and have begun predicting events, your organization will need to ensure it has the tools in place to integrate your predictive maintenance data with your existing technology. For example, you will want to integrate your predictive maintenance data with your ERP and/or CRM system to make sure the right part is brought to the right technician and they are dispatched at the most opportune time.
5. Service Management
With your systems integrated, you should be able to set up workflows which are kicked off by a triggered event. With workflows in place, deep analysis can be performed to improve and refine processes over time.
6. A Change Management Agreement
Ensure there is commitment and buy in from management. In order for any organizational or technological change to be effective, there needs to be company wide adoption. This is true of any large change to the way a business operates, and implementing a predictive maintenance program is no exception. By soliciting organizational commitment early and identifying executive champions you will be well on your way to ensuring user adoption and predictive maintenance success.
What Does a Predictive Maintenance Solution Look Like?
The way Hitachi Solutions approaches predictive maintenance is with our predictive maintenance reference architecture. In this architecture, Hitachi has developed a standardized dataflow/workflow for the quickest configuration of a multiplicity of IoT type devices and the data they generate. This data then feeds into an automated and self-improving maintenance solution, called the Hitachi Solutions IoT Predictive Service Hub.
At the highest level, this process starts with the registration and identification of new sensors that come online and get picked up by the Field Service hub. This hub will identify unique devices and the data associated with each device, whether it is temperature, humidity, electrical current, positioning, location data, etc. This data get streamed in and, based on a series of process rule trees, is stored or pulled into a field service workflow.
These workflows contain our first layer of proactive actions. Based off of the data contained within, we can trigger a variety of behaviors. Some of those may be to search for replacement inventory in our ERP system, or to trigger a reboot, an alarm, a device shutdown, or even to automatically schedule a maintenance appointment with the closest available service agent.
The second half of this architecture is the predictive element. Using machine learning, we are able to sweep through our data to predict when our machinery is going to fail and what type of a failure it will be, in real time or on a schedule. Furthermore, we can use machine learning to profile our devices for patterns of sensor readings that lead up to a failure. Then, as our machine learning processes identify these patterns, we integrate them as new rules into the proactive workflows to provide the customized rules for each device — allowing us to automate the management of each device with the utmost accuracy.
What Are the Considerations for a Predictive Maintenance Program?
The reality is that not every organization is ready for the leap into a predictive maintenance solution. Usually, this is due either to older equipment or to the fact that no historical sensor data has been stored. This presents us with a challenge. How are we supposed to identify when failures have occurred so we can train our prediction models to help us going forward?
While problems like these may certainly impact how soon you could have a predictive maintenance solution running, they may not exclude you from getting started. The first step would be to begin storing your data as soon as possible. In an organization that relies on machines to be operational, data about the state of the machine is enormously important. Treat your data as an asset, something that will continually give you better returns over time. Storage of very large quantities of data is very inexpensive today – take advantage of this.
The second step is to begin the process of identifying the key outputs you are trying to predict. Most frequently this would be the failure of a device, or the remaining time until a failure of a device or a component on a machine. You may not be capturing this information today, but, by examining your data, you may be able to create the data points that measure these key outputs (with some work and through the combination and processing of existing data). This gives you a starting point to execute your prediction process.
Finally, you don’t necessarily need perfect prediction results to begin leveraging machine learning for your decision making. For example, many machine learning outputs will include a probability of the predictions’ accuracy. In a case like this you can identify thresholds that a probability must exceed in order to be actioned upon.
As with any prediction, you will need to continue to retrain your predictive models, as the data that is available to you increases and your equipment changes. This means that you will always need to be cognizant of a margin of error that is inseparable from your predictions, but it also means perfection is not required to begin gaining value from recasting your business problems through a predictive lens.
The opportunities that we have to improve businesses on this significant a scale are too valuable to leave unexplored. IoT data has left this opportunity wide open for most organizations working with machinery or anything that captures sensor data. The obvious, yet enormously impactful, case of predictive maintenance is just the beginning of the potential we will see from the steady reduction in the cost of capturing data and calculating predictive outputs. We at Hitachi Solutions see the inexpensive cost of IoT and machine learning coming together to reshape how we ask many of our business questions today, across almost every industry.
Predictive maintenance is the most meaningful use case today, however, we are already seeing equivalent reconceptualization of business problems in the retail, warehousing and insurance industries. An exciting moment lies just before us!
Hitachi Solutions demonstrated the exciting features of the Hitachi Solutions IoT Predictive Service Hub at the virtual event, The Manufacturer of the Future: How IoT and Predictive Analytics Transform Operations for the Digital Age, Don’t miss out – click here to watch the recording!
For further questions on implementing a predictive maintenance program or for questions on the Hitachi Solutions IoT Predictive Service Hub, please contact Hitachi Solutions today.