I recently was requested to do a control performance audit in a mine situated in the Sahara Desert. It was an extremely interesting trip, apart from the fact that it also entailed nearly 3 days travel each way, with some long, tiring, and very tedious flights.

The mine is run mainly by expatriates coming from countries as diverse as South Africa, Australia, Northern America, Philippines, France, Germany, Britain, and a few more. Really multicultural. As a home loving person, I am always amazed by the life these expatriates lead, working typically nine weeks on site, and three weeks off back home. It is a tough life for both them and their families. When on site they typically work a 12 – 14 hour day, 7 days of the week.

I had always believed that the companies running these plants made conditions on them as attractive as possible to keep staff happy. This certainly wasn’t the case on this mine. Accommodation consists of a tiny “hutch” in a container, with a bed, a cupboard, a small table, and a few shelves, and pretty primitive communal ablution facilities. I wasn’t surprised that their staff turnover is extremely high. Not many people last longer than 18 months, and then the people move on to other plants in remote spots. It is certainly not the life style I would personally ever choose, even if the pay is good.

Typical day temperatures whilst I was there were in the 40º C range, but I was told it can get much warmer in mid-summer. Air-conditioning is a must. I really can’t imagine how people lived in the early days without it. Some days were very windy with brown dusty sand being blown everywhere. However one night I was woken up by the arrival of the most incredible storm. An extremely high wind which was followed by a torrential rain storm. This was the last thing I expected to encounter in the desert!

The storm lasted for about 4 hours and resulted in the most amazing flash flood. In the morning virtually the whole area was under water. Access to the mine was cut off by nearly 6 hours. Fences were washed away. I saw a caravan that had been smashed into an unrecognisable mess by the wind and water.

Enough of the travelogue, and back to the problems on the plant which had called me in because they were experiencing tremendous troubles with their controls.

The control system was based in a well known make of European PLC’s, which are commonly used by many mining groups. The following are extracts taken from my report:

The two important things that need to be done, particularly with control blocks on these PLC’s, are:

Each control block must be triggered by a repeatable accurately timed interrupt, and this same time must be entered in the cycle time of the block. (Typically a 1 second cycle time is good for industrial control for various reasons). Integral and derivative values which are time dependent will be wrong if this is not done.

The control block must have both the SP and PV values normalised to a range of 0-100% on the actual inputs to the block. If this is not done and they are left in engineering units with ranges not 0-100% the proportional gain in the controller will be wrong.

Neither of the above things has been done correctly in the general programming of the PLC’s, so that scientific tuning of most loops became very difficult, if not impossible. Even normal “trial and error” tuning which is based on people’s experience and on “feeling” becomes impossible as each loop’s controller can behave differently.

Normally one can investigate what has been done in the PLC programme and then apply corrections to each controller, based on common factors, for example taking the measurement span into an equation to compensate for controller gain values. Unfortunately in this case it was found that individual control loops were not working the same way, and that the factors differ. The programme is extremely complex and we were not able to determine what the original programmers had done to make such a mess of things.

System integrators normally use their own functional specification on programming control blocks into a system so that common routines and sub-routines would apply on all loops. Such a specification did not exist on the mine. Also to complicate matters even further, apparently many other people had subsequently worked on the programme so that not that much commonality still exists. So not only did controllers in different PLC’s operate differently, but it was found that controllers in the same PLC also could, and often did, operate differently from each other.

No anti-reset windup protection had been programmed into the controllers, which needs to be done externally on this type PLC’s control block. This can result in the integral winding-up into saturation if the PV cannot get to setpoint for any reason. It can in certain case take many minutes or sometimes even hours for the integral to “unwind” when the loop does come back into a controllable situation. It can also cause windup in certain cases when the output reaches a limit during normal control on setpoint or load changes. It was recommended that this be programmed in.

I recommended that even in spite of the cost, it might well pay the mine to get an expert programmer on that make of PLC to come on site to investigate and trouble shoot the programme, and if necessary rewrite control software, so that all controllers operate correctly in a common manner.

I have two interesting examples of control loop problems found at the site. The first is an airflow control loop where is fed into the bottom of a floatation tank and causes bubbles to rise in the slurry, carrying certain valuable components in the slurry with them. (Refer also to Case History 105 which discussed this process in much more detail.)

The airflow is very important, and in this case the mine was using an APC (advanced control system) to control the actual bubble size, which is measured by a special analyser. The APC then sets the setpoint of the airflow controller. APC setpoint signals like this coming from an APC system, are generally always moving around fairly slowly, and if the APC is to operate effectively it is essential that the process variable (the airflow in this case) follows the setpoint with absolutely minimum variance.

Figure 1 is the closed loop test, with the ‘as found’ tuning in the controller, and with step changes being made on the setpoint. (Note that the controller was in local setpoint, i.e. disconnected from the APC.)

Fig. 1

The following things can be clearly seen:

The valve is displaying absolutely huge hysteresis as the PD (controller output) has to move back through about 20%, before the valve starts reversing.

The measurement is at far too low a value to be accurate or even creditable. The transmitter in use was a thermal type flowmeter, but I was unable to find specifications at the plant on its rangeability. However it is unlikely any type of flow transmitter could give reliable, let alone accurate readings as low down as this.

The valve is also working far too close to seat. It is remarkable that it was even able to give any control at all. As a very general rule control valves should be operated above 20%.

The tuning was very bad with far too slow an integral for a flow loop.

Fig. 2

Figure 2 shows the final closed loop test which was performed with reasonable tuning values in the controller. This test is a wonderful example of the detrimental effects of hysteresis on closed loop control, with huge reversals being made by the PD under the influence of the integral action, every time the controller needs to move the valve in the opposite direction. The figure was deliberately presented showing the whole 0-100% range, in order to illustrate how close both the measurement and valve are to zero. It was recommended that the valve should be replaced with a smaller one, and that the transmitter be recalibrated with a much smaller span.

The second example concerns the problems the plant incurred in tuning the controller of an important level control of a tank.

The control was extremely unstable with the level moving in a large cycle, and with the output opening and closing the valve over a wide range. Weeks of trying to get the process into control were of no avail. The original tuning parameters that were found in the controller were:

P = 0.17, and I = 95 seconds/repeat.

Investigation showed that the proportional gain in the controller was correct, but the integral calculation was drastically wrong. Using largely “trial and error” tuning, excellent control was obtained with the following parameter values being inserted in the controller:

P = 15.0, and I = 3,200.0 seconds/repeat.

Comparison of the two tunings will give an insight into why the loop was previously so unstable!

A short while later the plant shut down and it was possible to do a test on the controller to actually establish the real value of the 3,200.0 seconds/repeat integral. It’s was in reality 290 seconds/repeat. How can PLC programmers get things so wrong?

Fig. 3

Figure 3 shows a section of the trend of this control loop taken from the plant’s SCADA, before and after tuning. It shows the remarkably good control of the level, and how quickly it managed to catch the two load upsets. (The pump on the outlet of the tank was switched off at one stage for a while which is why the controller’s output went to zero and the level up to the top.)

This example really shows how difficult it was for the plant’s C & I technicians o try and tune controllers, with the controllers all working in a different way. As mentioned in several previous articles I generally find at least 85% of controllers in PLC’s set up incorrectly! However these ones were about the worst I have seen.