Predicting maintenance for vehicle engines 2

What engines, what vehicles, what is considered "maintenance"?

The engine manufacturers have that information, and they're keeping it to themselves. I'm sure they are researching the same thing, so they have a big advantage over you. If you were somewhat knowledgeable about engines yourself, you wouldn't have to ask your question. Since evidently you're not particularly knowledgeable about engines, you really need a full time collaborator who is.
Failing that, my suggestion is, since data collection, storage and analysis is cheap these days, is collect and analyze data on every engine variable that could possibly have a bearing on maintenance; also collect data on the maintenance required/performed, and use 6-sigma tools to figure out which variables matter for which kind of maintenance. Don't forget to consider environmental factors and operating patterns.

"Schiefgehen wird, was schiefgehen kann" - das Murphygesetz

Step 1: Obtain the vehicle's owners' book.
Step 2: Find, within that book, the chart outlining the age and distance at which various checks are to be performed. If the vehicle had multiple choices of powertrains, ensure that you are following the chart for the correct one.
Step 3: Follow it.

If you are in an end-user position - Why is machine learning necessary for something like this? If you deviate from manufacturer's recommendations, particularly during the warranty period, and something unexpected happens, you may encounter difficulties obtaining warranty coverage. The only thing you need to "learn" is the manufacturer's recommended maintenance schedule.

If you are in an OEM position or perhaps a fleet operator - Many vehicles nowadays have somewhat-automated maintenance reminders. Are you trying to train something like that? Most of them just count off calendar time and distance. Some of them might also count off cold-start cycles or running time. A few of them may account for events such as distance or time above a particular RPM threshold, or above or below particular oil or coolant temperatures, indicative of severe-service operation (hard driving, trailer-tow, short-trip operation, operation in severe hot or cold weather). Time or distance driven in fault conditions ("check-engine" fault warning lamp on due to misfire or air-fuel-ratio abnormality) might be an interesting factor, too.

Some engines have timing belts needing periodic replacement. Petrol engines have sparkplugs needing periodic replacement. Air filters need periodic replacement but how is your fancy machine-learning system going to know if the car drives through a sandstorm, or a forest fire, or a volcanic eruption? Belts need periodic replacement. Coolant needs periodic replacement. Lots and lots of other stuff nowadays needs no scheduled maintenance but goes in the "fix when broken" category.

How is your machine-learning system going to know if the operator used the correct grade of fancy synthetic oil meeting a particular tight specification ... or the cheapest stuff they could find that said "oil" on the bottle?

Same, but substitute "coolant" for "oil"?

There's a lot of differences in both powertrains and applications that I have my doubts! What's appropriate for a smaller engine having 3 litres of oil in the sump might not be appropriate for something having 15 litres in the sump.

What are you going to use for feedback that your machine-learning has learned the correct things?

Valve recession IS the key wear factor for modern engines.

An engine won't see valve recession if it burns most of the oil between manufacturer's recommended changes and the oil light isn't a reminder to put more oil in, but that you've already run out.

I don't recall which off hand, but some passenger vehicle engine is infamous for the oil rings getting stuck, followed by too much oil getting by, which would make a puff of blue smoke except it gets caught and burned off in the catalytic converter. The owner only finds out how bad this is when the "engine scoring" light comes on.


Though not nearly as much fun are the extended ceramic plugs with close fit in the head (to limit non-combustion volume) that allow carbon to fill the gap and seize the plugs over the extended life change suggestion by the maker. (Triton V10) I think the plug changes for a time were $100-$200 for each plug to compensate the risk of having to remove the head to retrieve the jammed plug fragment. I think that interval went from 100,000 miles to 30,000 miles to avoid the problem, at least the interval the owners used.

The plugs that would not come out replaced the design that sometimes coughed the plugs out. Progress? Helicoils were easier to put in than getting that plug fragment out.

Maintenance can be a crap shoot when it comes to predictions.

As a fleet operator of high horsepower diesel engines (1600-3400hp), cylinder heads are the short life item on the engines. We replace them at half of the cylinder/piston life. A dropped valve due to recession is the most destructive result of wear

Having no oil isn't a problem? I want an oil-free engine in my car.

Valve recession may be the key wear factor for SOME modern engines, but there's lots of modern engines with fixed (shim-and-bucket) valve lifters that go 400 000+ km without being touched.

3DDave said:

Having no oil isn't a problem? I want an oil-free engine in my car.

Click to expand...

The motor + gear-reducer + final drive assembly in my Chevy Bolt contains about half a litre of oil that is supposed to be changed at 150 000 km ...

Not sure which aspect of predictive analytics you're researching, but there are quite a few SAE papers on engine maintenance applications and some of this has long been in production. ~20 years ago while still a diesel tech I worked on failure models for FEAD and various accessory components, and have managed folks working on oil and filter life models.

Looks like a fools errand to me.

Maintenance periods and activities are simply the mean point between excessive wear and failure of the engine before its allotted lifespan versus cost of the said maintenance. The OEM is the only person who can decide where that sweet spot is.

Set it too high (and miles or km are the only thing most people understand) and your engine fails prematurely and they get a bad reputation and people sue you. Set it too low and your engine gets a bad reputation for costing more than the others to keep it going.

Ditto all the systems which supposedly track the things Brian listed need a huge amount of research and actual engine strip down data to know how any of those items actually affects the life of the engine so that you can start to modify your default of 10,000 miles or whatever.

Only the OEM knows these things and even then they are probably limited on how many engines they destroy to find out what happens if you run the engine at 10% of its power a lot of the time or 95%. Or have 100 stop starts in a day compared to 5 or less.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

I had a customer that builds automotive belt-tensioners, idler pulleys, alternator over-running clutches, and the like. They have a room full of test stations running 24/7 which drive pulleys and belt-drives in a way that simulates the motion profile of the engine that they're meant to be attached to, and they're each inside temperature-controlled sealed enclosures that they can use to emulate practically any foreseeable weather condition including being sprayed with dirty and/or salty water.

At the time, I had a vehicle equipped with a timing-belt tensioner that they built. When the time came for a timing-belt job (which I did myself on that car), I offered to supply them the take-off parts for analysis, and in return, they supplied me a new replacement tensioner straight off the production line.

I highly doubt that their test data is shared with anyone other than the vehicle manufacturer that they are building a specific part for, and only the test data for that part.

Some modern engines use electronic control of oil pressure. By monitoring actual oil pressure against input signal to the pressure control actuator while taking into account oil temperature, time since last oil change, engine speed, and other factors, it may be possible to establish a relationship that the bearings are excessively worn because the control valve is having to return less and less oil back to the sump and divert more of the oil pumps displacement to the bearings.

Both the OEs and suppliers are amazingly open to sharing data and SMEs if you have a decent business case and budget, and there's some interestingly diverse business cases affecting the use of analytics.

Predicting every component's failure is a great idea in theory but in application the question is, "I have data suggesting failure may occur, now what?" In the typical light-duty passenger application, most err on the side of caution with potential core/long-block problems and shut the engine down bc catastrophic failures are an expensive PR nightmare for both consumers and OEs. Accessories like alternators OTOH are typically allowed to run themselves to death bc replacement is cheap, analytics add cost, failures aren't catastrophic, and popping a MIL/code for a component which likely still bench-tests fine causes maintenance and quality/perception concerns.

Some business cases like military vehicles and aircraft can be the exact opposite, and rely heavily on analytics. When lives depend on an engine running, drivers generally want to identify every possible failure including cheap accessories before leaving base/garage so they can complete repairs or use a different vehicle. Once out on the road, they often want low-oil and other core safeties overriden to ensure the vehicle keeps running as long as possible and even the core engine is essentially disposable.

Looks like our OP is taking us for an April fool....

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

Could be, but not a good idea in an international forum when all time zones are in play. My timestamp for the OP shows 1 Apr 24 23:17.

"Schiefgehen wird, was schiefgehen kann" - das Murphygesetz