An expert way to identify faults

Most large steel or chemical plants across the UK now employ computer-based SCADA packages that allow operators to measure and control the processes within the plant very accurately. What they do not do so well is to help plant management in the field of

To answer these questions requires a system specifically tailored to predictive maintenance. Now, thanks to work carried out by Ken Gale at the University of Cardiff and commercialised by his company Expert Monitoring, the industry has a solution an expert solution at that.

Ken Gale’s system is based on a simple premise. He felt that if the operation of individual components within a hydraulic system and the interaction of these components within the whole system could be understood thoroughly, then that knowledge could be encapsulated in a computer system. Rules could then be written in an expert system programming language that would make inferences on that data when it was presented to it. These inferences could then predict, for example, if a solenoid valve was likely to stick, or a hydraulic motor was close to failure. And that, in a nutshell, is what Expert Monitoring has done.

Making measurements is key to the understanding of how components and complete systems work. In one instance, Expert Monitoring worked with a major steel company to determine the cause of vibration on a mill stand. The system under examination comprised a hydraulic servo valve connected to a number of roll bending rams.

The technique that Expert Monitoring built to measure vibration in the control system was itself highly innovative. The monitoring hardware actually connected directly into the live analogue signals of the input to the servo valve through the use of an opto-isolator. From there, data in the form of electrical signals was converted to digital form and then fed into a PC.

Expert Monitoring used the Labview software package from National Instruments to help with the collection and signal conditioning of the data. The data was then stored and used by Expert Monitoring’s own custom software to determine what sorts of faults might be present in the system. ‘It is possible to spot faults by this method because any disturbances in either the two stage action of the servo valve or the alignment of the roll stack assembly can be detected by the application of a Fast Fourier Transform (FFT) on the measured line voltage from the transducer,’ says Gale.

In other words, because Expert Monitoring set up a direct connection to the control system, it was possible to determine what sort of faults were occurring simply by looking at the frequency spectrum induced by the fault: a signal below 45Hz is likely to be a mechanical fault, a signal between 50 and 100Hz is likely to be an electrical fault, and any signal above 320Hz is likely to be an electronic fault.

Albeit the fact that the system was effective in diagnosing faults, the company wanted to take the system one stage further, so that all the knowledge relating to closed loop control, both mechanical and electrical could be built into an expert system as rules that could then be used for fault finding.

To complement the information that is collected by Expert Monitoring, the system can be interfaced to existing PLCs that may also be collecting data that could potentially be useful in fault diagnostics.

‘In an existing control system, PLCs are also in the business of identifying faults. We have written a software driver to our system that allows us to pull all of this information out and use it too,’ says Gale.

Once the data has been collected from the system, relevant parts of it are sent to an SQL database that interfaces directly to a Gensym G2 system. The company has used the Gensym G2 system to allow it to write a fairly straightforward set of ‘rules’ that are then used to interpret the data and make decisions based on it.

G2’s rules work in real time and can mimic the human ability to focus on specific problems. They can be event-driven (through forward chaining) to automatically respond whenever new data arrives.

‘There are a number of effects associated with the failure of a solenoid, for example,’ says Gale. ‘You can use Gensym to describe those effects using the rule based system. In that way, if the system ‘sees’ those effects through the measurement of the back EMF through the opto-isolated coupling, it will be able to interpret the effects as a solenoid failure and present the result to the user.’

Some of the faults are very specific. ‘It was easy in some instances to relate a particular frequency to the occurrence of a certain type of failure,’ says Gale. That was the case with determining vibration problems, for example. ‘Other faults,’ he adds, ‘like offset faults, required a little more work.’ In such instances, there are a number of conditions that could occur to produce the same effect. With such faults, the system first diagnoses the location of the failure at the transducer and then instructs a user to recalibrate a particular system using a Fluke Process Calibrator.

A custom built user interface can show various levels of graphical detail to the operator, depending on what is required. At the highest level, a complete overview of the operation of the plant can be viewed, much in the same way that a modern SCADA system does. But the system can be interrogated further to show more intricate details at the component level too, highlighting a particular component failure.