Predictive maintenance (PdM) is all the jazz right now.
And for good reason:
The rise of Big Data, IoT, Cloud and AI is driving PdM with a 40% CAGR towards $10+ bn by 2022.
And the hype is real.
As evident below, Google Trends shows a 400%+ increase in interest since 2008:
Yet, with the hype also comes a ton of fluff & buzz articles.
In this article, we cut the clutter with a simple guide to PdM - with focus on CAN bus applications (vehicles & machines).
In particular, we outline what predictive maintenance is, how it works and why it’s attractive.
We end with practical examples and actionable next steps - so that you can stop reading and actually start a pilot.
WHAT IS PREDICTIVE MAINTENANCE, REALLY?
Let’s start with a really basic definition:
The goal of predictive maintenance
is to forecast & prevent equipment failure
Let’s try and put it in the context of two classic maintenance methods:
REACTIVE maintenance is simple: “If it’s not broken, don’t fix it”. It’s best for cases where equipment failure is rare, easy to fix and with limited impact - e.g. switching a lightbulb in a warehouse
PLANNED maintenance (or preventive maintenance) uses routine maintenance to diagnose equipment for failure. It works and is widely employed - but it’s costly and doesn’t capture asset-specific conditions
PREDICTIVE maintenance leverages data from an individual assets to predict failure. This way, repairs can be done when needed (and avoided when not). It offers the best upside, but at the cost of complexity
Of course, predictive maintenance can be further split into sub sections - see details below:
Predictive Maintenance 1.0 to 4.0 - Details
In principle, predictive maintenance has been around for ages: When a technician inspects an asset and makes a change to avoid future failure, that’s predictive maintenance.
The difference is the amount of data that is used and the frequency of updates - cf. the overview of PdM 1.0-4.0.
Below, we focus mainly on Predictive Maintenance 4.0, using IoT, big data and predictive analytics in near real-time.
In particular, we'll dive into the sub sections of PdM 4.0 to show that this is still an extremely broad segmentation.
On the surface, predictive maintenance is kind of simple.
Yet, in a 2017 survey of 280 companies PwC found that only 11% have reached “PdM 4.0” maturity (i.e. using big data to do predictive analytics) - though 47% plan to implement it in the future.
That’s a big disconnect!
If PdM is simple, then why haven’t more companies already implemented it?
Let’s explore this below.
HOW DOES PREDICTIVE MAINTENANCE WORK?
You can look at predictive maintenance as a process flow with 3 steps - as illustrated below:
1# COLLECT: First, you’ll need to gather relevant data that can help you predict time-to-failure. This is often done by e.g. adding vibration IoT sensors to get “indirect” data.
However, a far more direct method is to tap into the run-time data the machine uses to operate. All vehicles - and most industrial machinery - operate using a combination of transducers (sensors and actuators) and control units. The control units implement algorithms that determine the behavior of the machine. The run-time data is continuously transported, often using wires and communication protocols.
One very popular transportation method is "CAN bus", whihch is e.g. standard in all vehicles. If a vehicle or machine uses CAN bus wires for data transportation, it's often possible to safely tap directly into the run-time data stream. This can be done using an IoT CAN bus data logger, which opens up the full scope of your asset operational data. Your CAN logger can then transfer this data to the cloud via a WiFi hotspot (WLAN, 3G, 4G) in near real-time.
2# PREDICT: The collected data is processed in the cloud. For CAN bus data, this includes transforming the data to scaled engineering values. Once ready, the data can be used in a predictive model. These range from simplistic single-variable thresholds - to advanced machine learning algorithms. More on this below.
3# REACT: The model provides estimates of e.g. the time-to-failure for an asset and its components. From here, it’s “simply” reacting on the insight: Auto-schedule maintenance, send push notifications to warn staff of potential breakdowns and optimize your spare part inventory.
PRACTICAL PREDICTION MODEL EXAMPLES
Do you feel the PREDICT step is a bit fuzzy? You’re not alone!
Below we try and flesh this out with some very basic illustrative examples:
Example 1# - The Oil-Sensitive Truck: In this Google Sheets illustrative example we consider a vehicle where the only possible failure is from not refilling the oil. Over time, the oil level decreases to 10% and the vehicle breaks down - its ‘Remaining Useful Life’ (RUL) is zero. In the example, we pretend that we observe the true RUL and adapt a simple linear regression. With this we predict the RUL out-of-sample and decide a time for warning - which in the example is ~17% off vs. the true RUL.
The point here is merely to show the concept in a simple way - feel free to copy & try it out yourself!
Example 2# - NASA Turbofan Engine: We strongly recommend reading this great practical intro by Ben Everson. In short, he takes outset in NASA’s 2008 Turbofan Engine Degradation simulation data set), which contains training & test data on 10 engines equipped with 20+ sensors. The goal is similar to Example 1#, but in a more realistic context. In particular, this is helpful if you're about to get started with the basic data cleaning & statistical considerations.
Check out both his intro and follow-up!
The above examples illustrate that adding more data quickly adds complexity.
For this reason, PdM using machine learning (ML) has become popular in predictive models.
Machine learning is a natural fit as PdM involves failure classification using huge amounts of sensor data. We won’t dive deeper into PdM using machine learning here, but consider reading this InfoQ technical intro or this Azure AI article using the NASA Turbofan data with ML.
Further, if you’re interested in a bit more detail on the predictive models for PdM, check out this article by BigData Republic - it’s a great read, in particular to provide context for the RUL modelling strategy in our examples 1# and 2# above.
We’ll detail how to get started further below - but first we’ll look at the potential benefits of PdM:
IMPACT & BENEFITS OF PREDICTIVE MAINTENANCE
If you succeed with PdM, it can be game changing - with examples of impact statistics including:
Source 1: 30-50% | Source 2: 10-20% | Source 3: 70%
Below we detail some of the major benefits:
BOOST DELIVERY ACCURACY: If there’s one thing your customers hate, it’s poor delivery accuracy. By employing PdM, you can drastically reduce unexpected breakdowns in your vehicle fleet or site machinery. With demanding players like Amazon, this can be a strategic make-or-break for many big manufacturers.
MINIMIZE ASSET DOWNTIME: By fixing issues before they lead to breakdowns, you’ll immediately reduce downtime. Further, a CAN bus PdM system provides your technicians with extensive asset-specific data. This way, they’ll know the issue beforehand - letting them skip generic checklists and cut downtime dramatically
REDUCE MAINTENANCE COSTS: Preventive maintenance is costly as it relies on e.g. time based intervals. These are weak predictors of failure, hence conservatively frequent check-ins are needed. With PdM, the asset-specific big data is far more precise in predicting an issue - enabling far less frequent maintenance
OPTIMIZE SPARE PARTS INVENTORY: With reactive or preventive maintenance, a massive inventory of spare parts is necessary - you don’t know what you’ll need until the asset is taken out of operation. With PdM, you’ll estimate time-to-failure for each sub component via the cloud, allowing a far more lean inventory
CUT FUEL COSTS & EXTEND LIFE: By setting up PdM for e.g. a vehicle fleet, you can track the remaining life of each vehicle component. This lets you truly optimize total cost of ownership (TCO), by deciding the optimal replacement timing taking into account replacement costs vs. impacts on fuel costs and vehicle life
IMPROVE WORKER SAFETY: In worst case, breakdowns of machinery or vehicles can lead to catastrophic events and harm workers. By predicting issues before they escalate, you’ll be able to reduce accidents and boost team morale
GO BEYOND: Finally, basing your PdM on "direct data" is far more powerful than e.g. using vibration sensors. With CAN bus data, you’ll practically have a digital copy of your asset in the cloud - opening up tons of new applications: Productivity optimization, remote troubleshooting, real-time KPI dashboards and more. In turn, this improves the return-on-investment (ROI) on your implementation.
The benefits can be harvested by both end users (e.g. site or fleet managers) and original equipment manufacturers (OEMs).
Furter, PdM is highly relevant across both heavy duty vehicle fleets (trucks, buses), defense vehicles, agricultural equipment, industrial production machinery, battery management - and many more industries.
In short: With predictive maintenance, the sky’s the limit!
But of course it’s not that easy - in particular, five major challenges come into play:
Five Key Challenges in Implementing Predictive Maintenance
Changing your culture - PdM is unknown territory to most corporate leaders. “Will this work?”, “Does it kill our budget?” - get buy-in (or do light PoCs under the radar)
Choosing the right IoT solution - the market is complex and finding a good fit is hard. If you’re just getting started with PdM, avoid an expensive enterprise solution. Start small
Managing big data - processing big data can be difficult. Set the right team and begin with few parameters. But: Historical data is vital - so store all relevant data you can
Avoiding data lock-in - most IoT solutions lock your data in the their cloud. However, this data and the capabilities learned may become your top differentiators. Own your solution
Building your model - the model should match your capabilities. Don’t choose a big outsourced model you don't understand if it’s not a fit. Start simple
The challenges are significant - and implementing PdM requires the right approach.
So how do you get started?
We explore this in our next section!
STEP-BY-STEP: GET STARTED WITH PREDICTIVE MAINTENANCE
Back to the question:
Why have so few corporates implemented predictive maintenance?
From our experience, the answer is pretty simple:
Too much talk. Too little action.
PdM gets plenty of hype even at the top - but dies because nobody get started.
So how do you do it?
The key is to do a small, simple and low cost proof-of-concept (PoC)
- that scales easily
Why? When you bring the PoC to your management team, you drastically reduce the need for them to “believe”.
Instead they can see. This opens up the budgets for scaling - so you can move to the next phase.
To do a PoC for predictive maintenance, our advice is to follow the below steps:
CHEAT SHEET: PREDICTIVE MAINTENANCE PROOF OF CONCEPT
1# Identify a well-known pain point - is Machine X at Site Y a well-known problem child? Then that’s a good place to start. If the machine operates on CAN bus, it’s a great candidate for your PdM PoC
2# Set your team - fully dedicate a 3-4 person team to implement the PoC over 3-6 months. Ideally get an application expert, data expert, coder and a communicator (to sell results internally)
3# Define your predictive model - identify what failure type you’ll predict, what the failure process looks like (fast, slow, …), how you’ll predict it and what data is required
4# Acquire 1-10 simple-to-use IoT data loggers - ideally limit spending to 500-4000€ max. Check with the supplier that the loggers are able to record data from your application if in doubt
5# Collect all relevant data to your cloud - connect the IoT CAN loggers to your asset and a WiFi hotspot to set up the transfer to your cloud server. Start building your data processing scripts
6# “Train” your predictive model - as you get historical data (incl. failure data), you can adapt your model parameters to improve your model’s ability to forecast out-of-sample failure
7# Move your PdM model from “development” to “production” - with your trained model, you can now actually start monitoring if your model is able to identify unexpected issues early on
8# Add reaction engine - to “sell” the PoC to management, we recommend incorporating simple auto-reactions - e.g. sending an SMS to a site manager when an issue is identified
9# “Sell”, develop & scale - present the PoC, results and a business case / roadmap for scaling it up. The purpose is to get management on-board to provide funding for the next phase
Note two subtle points:
First, you ideally need (a lot of) regular data & failure data to train your predictive model - so get started ASAP.
- Further, make sure to take the “data gathering period” into account in your planning
- Alternatively, you can invest in more devices to speed things up if you have the budget (and enough identical assets)
Second, it’s vital that you don’t drop the ball during step 9#:
- Avoid over-promising in your business case and demanding absurd budgets
- Avoid trying to implement PdM across everything in year one
- Avoid shifting your PoC model to an advanced machine learning model day 1 (even if your CEO says so)
You need to match expectations vs. capabilities and take your time.
This is also why you benefit from a zero-subscription solution where you own the data. This way you avoid a continuous cash-drain (i.e. time pressure) and you have the basis for developing in-house digital capabilities and a custom-fit platform.
NEXT STEPS: CHOOSE AN IOT LOGGER
As mentioned, you’ll want to start data collection asap - hence you need an IoT data logger, e.g. the CL3000:
CL3000: Low Cost IoT CAN Bus Data Logger
If you aim to use CAN bus data as your basis, we strongly recommend the CL3000 CAN bus data logger with WiFi. This device has a number of advantages for predictive maintenance:
Low Cost & Unlocked - The CL3000 costs only 399 EUR/unit with free software, free shipping and no subscription fees. This is lowest cost IoT CAN logger available - perfect for getting started and scaling up
Plug & Play - Simply connect the CL3000 and it’ll start recording data to the SD card. Further, you can configure it to send this via a WiFi hotspot to your FTP cloud server in less than 5 minutes
Easily Transfer Data - Once connected, the logger will simply push log files to your server in e.g. 1 MB chunks - ready for being auto-converted to readable form by our free software CANvas
Supports J1939, CANopen & More - The CL3000 works on all high-speed CAN bus systems, incl. J1939 (e.g. trucks, buses, cranes, excavators, …), OBD2 and CANopen (e.g. production machinery)
Want to know more? Read our detailed intro to the CL3000
"What impressed me most was the responsive and helpful support of CSS:
They really listen to their customers, suggestions for improvements were heard!"
- Harald H, B-sab GmbH
We hope you found this article useful!
At CSS, we love working with our end users - and we’d be happy to spar with you to find an ideal solution!
Send us a mail and we’ll get back to you asap!
Liked this article? Please share!
RECOMMENDED FOR YOU