Skip to main content

Machine Maintenance: If only they could talk

Robot_factory1_748436305a2422c4dff89277f0f6a3a4734ac892.jpgAn essential feature of any automated factory based on the concepts of Industry 4.0 is the ability to predict mechanical failure of the machines and if not fix faults automatically, at least provide a timely warning to the guys with the spanners.  Image: Wikipedia

The Decline of the Machines?

What does the word ‘machine’ mean to you? If your experience of ‘technology’ is limited to the last decade or so, then some sort of computer probably springs to mind, based on solid-state electronic components. Being old and decrepit, I think of mechanical or electro-mechanical devices in factories; oily, noisy things perhaps turning blocks of steel into car components. Nowadays we talk of Machine Learning: the machine, in this case, being a piece of programmable electronics. The industrial era of vast factories full of skilled workers machining bits of metal using manually-operated lathes and milling machines is fast becoming a distant memory. As far back as the 1940s, Numerical Control was becoming common – people ousted by punched-tape, closely followed by Computer Numerical Control or CNC. With all this talk of computer control, automated factories, Industry 4.0 and the like, we can get carried away thinking about the replacement of human with artificial skills and ignore the mechanical devices that actually do the work. It’s very easy to forget that under all that sophisticated electronics lie machines that cut and shape materials into products, or parts of products using age-old techniques. With one big exception: 3D Printing or Additive Manufacturing may in time replace the old Subtractive methods. So, is the end of noisy machines with big electric motors in sight? Hardly: 3D printers are mechanical devices and the industrial-sized ones will still have big motors, gearboxes, rotating shafts, bearings….. In other words, no, machinery will be around or a long time yet.

But mechanical systems are unreliable and wear out…

Well, that depends how well designed they are to withstand overloads and abuse - something that should not be a problem with human operators out of the picture – and if condition monitoring and regular maintenance are carried out. However, don’t get the idea that electronic systems have an infinite life – see my post On Whiskers and Dendrites. Digital electronic circuits tend to work perfectly and then suddenly stop without warning. Passive components such as resistors and capacitors do age; their values changing gradually with time. But these changes can go unnoticed in a well-designed analogue audio amplifier circuit for example, until one day you compare its bass response with a new unit and realise that the electrolytic capacitors on the speaker outputs have ‘dried-out’! The amplifier still basically ‘worked’. The problem is: how can you monitor component aging in an analogue circuit in order to predict a future unacceptable drop in performance? A digital circuit can age too, but because it works with just two voltage levels which can vary a bit as long as they stay away from a threshold value, optimal performance continues. This makes it just as difficult to create an advanced failure warning system for a digital system.

On the other hand, the health of mechanical machines with rotating components is a lot easier to read. They generate warning signals indicating future component failure perhaps months in advance – if you know how to interpret them. Electronic circuits are silent and inscrutable, mechanical systems make a noise, even when new. A bearing just beginning to wear may produce inaudible high-frequency ultrasound. As it gets near failure – seizing or breaking-up – sound turns to a vibration you can feel and hear. The aim is to detect and fix the problem well before the vibration phase.

Listen to the sound, feel the vibration

You will not be surprised to hear that diagnosing faults on machines by listening to them running is not a new idea. Experienced car mechanics once used a powerful diagnostic tool – a long screwdriver with the handle against an ear and the tip touching various points on the engine block – to identify worn-out bearings and other moving parts. A kind of stethoscope for machine ailments. The screwdriver and the mechanic’s ear have been replaced by MEMS technology in the form of a specialised 3-axis accelerometer chip or Vibrometer, for example, the STM IIS3DWB (201-0404) . If you think about it, nearly all machines contain rotating parts and this means any noise they produce will contain peaks in its frequency spectrum corresponding to those rotational shaft speeds. Minor defects in bearing surfaces will contribute more peaks. All of this adds up to a unique ‘signature’; if recorded when the machine is new, it will act as a reference for comparison purposes at service intervals.

The vibration sensor has replaced the mechanic’s ear and screwdriver, but not his or her brain. The sensor output consists of time-sampled acceleration values in three orthogonal directions. A Fourier Transform performed on this data will reveal the frequency spectrum of the vibration signal. Unfortunately, in practice, it can be very difficult to extract the subtle failure data from all the random noise present in a signal coming from a bearing in the very early stages of wearing-out. By the time the tell-tale frequencies become detectable, it may be rattling around enough to be heard! Many researchers are working on new techniques for Feature Extraction in the area of Early Fault Diagnosis, and inevitably Artificial Intelligence is becoming popular [1].

Closing the loop

In a fully-automated factory, EFD is vital to avoid expensive downtime, but it’s somewhat sub-optimal having to call in human repair-people to ‘close-the-loop’, that is, fix the fault. We are a long way off practical repair robots able to tackle all types of fault, but there are ways an intelligent control system can ensure maintenance is carried out at a convenient time:

  • Reduce the wear rate on the affected part by reducing the stress on it. This might be achieved by increasing lubrication, reducing the shaft speed or reducing the load. An ‘investigation’ by the controller might reveal that one of those is actually out of tolerance for some reason and in that case fix the problem without further intervention.
  • In critical situations, switch in a spare machine to replace the faulty one. Obviously, this requires a lot of up-front investment and the provision of such cold spares, with the extra switching hardware necessary, would only be considered in cases where a failure could be catastrophic. The core cooling system in a nuclear reactor comes to mind.


An interesting example of ‘fault’ detection and automatic ‘repair’ that’s been around for many years is ‘knock’ detection and elimination for petrol engines. Engine knock or ‘pinking’ occurs when the ignition spark is triggered too early in the cycle and something like an explosion happens as opposed to a rapid but smooth expansion of hot gas in a cylinder. Modern engines feature a knock sensor which ‘hears’ the engine noise and a microprocessor which analyses the signal and alters the ignition timing if the onset of knocking is detected. This was an early application of real-time digital signal processing with DSP chips [2].


Machines with moving components can talk to us: it’s just a question of understanding their language. If the automated factories of Industry 4.0 are to be successful then the question of automated maintenance and repair will need to be addressed.


[1] A Review of Early Fault Diagnosis Approaches and their Applications in Rotating Machinery Entropy 2019

[2] Engine Knock Detection using Spectral Analysis Techniques with a TMS320 DSP Application Report SPRA039

If you're stuck for something to do, follow my posts on Twitter. I link to interesting articles on new electronics and related technologies, retweeting posts I spot about robots, space exploration and other issues.

Engineer, PhD, lecturer, freelance technical writer, blogger & tweeter interested in robots, AI, planetary explorers and all things electronic. STEM ambassador. Designed, built and programmed my first microcomputer in 1976. Still learning, still building, still coding today.
DesignSpark Electrical Logolinkedin