Predicting machine performance and reliability is a reality now. With rigorous methods of inspection, advanced instrumentation in place, machine reliability is already within industry’s reach. In a complex process plant, however, in my opinion machine reliability should not be focused alone as a distinct subject matter. Machine is used to perform the process operations; for obvious reason therefore, performance of process is also affecting a machine’s performance. If we focus only on machine’s reliability, I believe the subject is not complete. So, continuous process analysis and prediction of future where the process is heading to should be equally focused to achieve an ultimate reliability of the plant operations. In brief, if we have to achieve a reliable plant, developing a means of predicting the process and operations would be a great benefit to ensure an ultimate reliable and safe operations.
Cement plant operations is a complex process operation. It may sound odd to many, as vastly heavy mechanical mindset oriented industry may see it in as simple way as “breaking rock to making rock” type of operations. Comparing to intertwined and miles of piping and ducting work and distillation columns in a refinery industry, cement plant looks pretty straight forward except the fact that they look large! However, the fact is, very little we know about the complex chemical and mineralogical transformation that takes place in a Clinkerization process. If kiln goes upset, we crank up fuel or adjust fan or fuel or so forth, but very little we think or sometimes even know the true root cause why the kiln went to an upset condition.
Old timers used to burn the kiln using the empirical knowledge which worked pretty well in a wet process. They used to utilize their “feel” by looking into the flame color, burning zone’s appearance and look of charge at the burning zone by standing on the burner platform and peeping inside the kiln. That worked for a kiln which had a long retention time. With time, like everything changed, cement process has changed too. In a modern dry Precalcining cement kiln process, if things start going upset, operators do not get ample time to turn around the situation like old timers could do. On top of it, cement kiln never remained just a Clinkerization process anymore. With time, we learnt that it is an excellent incinerator that can destroy 99.9% hazardous organic waste that come from other industries. So, we figured out there is a fuel incentive if we can burn all the “trash” from the other industries, on top of it, it is a great contribution by the cement plants to keep a greener world. With the challenging fuels introduced in the process, a whole bunch of new tough regulations have been imposed to cement kilns just to make sure that what we are promising to burn, we really can burn it. So, with the challenging raw materials, fuels, alternative fuels, fast process dynamics and above and all a series of rigorous permits and compliances make the modern dry calcining cement plant operations way more difficult than burning wet kilns. Kiln operations has to be way smarter than what we think or get used to think now a day.
I believe just an automatic process control with some PID’s here and there and some “expert” control system in place which takes input from some traditional field instruments and based on this making a predictive decision is just not enough. Or in other way to say, it is just not smart enough considering to the risk and money that cement plant manufacturer’s put into the business.
To make a modern cement plant process really smart, I would envision the below:
- A parallel brain which will have cognitive ability: Yes, I am talking about what we call now a day as “artificial intelligence”. It will run in parallel all the “what-if logic “and “fault tree” possibilities that the processor will have it’s smart brain to analyze the system 24/7 and come up with messages in the screen like” Go, check your stage IV diverter gate towards calciner, may be you are getting more towards calciner that you believe” things like that. Before I give more specific example on this matter, let us try to understand the industry’s situation now. The so called “Expert” Control Systems can predict when things are running well. Meaning, it will rely on the field instruments (temperature, pressure, flow, vibration etc.) data given to the processor’s brain and provided that all are reporting well, the expert control system will take some action based on MPC ( Model Predictive Control ). In my humble opinion, this is an inferior grade smartness. We need superior grade smartness. We, the human beings, have to take pride that we are the superior smart brains, but we can’t work 24/7 and all people will not be same in analytical and intellectual ability who are running the kiln 24/7, so a new variance like “people skill” is now in front of us. Let’s take one example:
A cement plant is running with “Expert” Control System. It is a modern precalcining cement plant process where about 60% or above fuel input is by alternative fuels. Stack flow is strictly controlled, so limit has been imposed that the plant can’t go beyond so and so much of ACFM when the raw mill is in operation. In one winter night, the plant started showing a high stack flow ( if this exceeds, the alternate fuel will be off and plant management will not be happy to see that as the plant makes money by burning each tons of alternative fuel ). At the same time, the raw mill shows to be going colder, mill exit temperature is constantly dropping. So, to keep the mill hot, operator is drafting harder from the kiln to supply enough hot gas to the mill. This now creates the tower profile to be elevated further and main bag house temperature started to climb (which also an interlock as well fuel cut off exceedance ). So, to counteract all these, operator is cutting back the kiln feed, not to exceed the stack flow and also not to exceed the bag house temperature. Tower also shows to be little bit unstable as tower exit CO spike is present in every now and then.
In the above example, as what happens, problem elevates up to all “Key” people of the plant and they get text message and calls or mails and now it is up to them to wake up from sleep or stop the car in the road side to think, analyze and tell the operators to do something to try or tell them to wait until morning. This is the reality.
In my opinion, the system should be such that, it will give a jump start by providing some decisions for any abnormal issue that the process is facing at any given time. Also, it will be a watch dog of the “Expert” Control system and in parallel will run the “Fault Logic” to make sure math does add up and things make sense.
In this example, in troubleshooting this couple of things were found in the next day. The interesting one was, the raw mill exit temperature thermocouple was deceiving the control room. So, it was showing less than what the true temperature was. So, raw mill was actually hot, but as the operator was deceived and “Expert” control system also believed that what it is sensing was “right”, so it was also agreeing with the operator’s decision of drafting the tower harder or shutting the GCT water spray to give raw mill a “warm blanket” in a winter night. Poor baghouse started getting hotter and hotter as the bag house inlet temperature was not deceiving, thanks goodness. Now fresh air damper opens to cool down the baghouse, that exaggerates the stack flow further, so now operator cuts back the kiln feed any minimum feed up to what the stack flow allows him to be in operations. How about the tower CO spike then? It was probably a partial jam in the riser, so flow was hindered and material was flushing on and off. Moving the diverter and cardox blasting the suspected tower location helped to regain the tower stability to some extent.
Question is now can a so called “Expert” Process Control System analyze this situation and give some options or predictive decisions for the on-shift control room operators? The answer is NO. Can we do it? The answer is YES, all the technologies are available to make it happen.
- A Process Simulation which will run parallel to the main process:
All major OEM’s have a process simulation software to design the system. But what about every cement plant is having a process simulation running in parallel and giving the “Parallel brain” a cognitive support and helping it to take decisions. In this above example, if my simulation had to run the raw mill real time heat balance based on the inputs that the system is getting, don’t we think that it would be a piece of cake for the “Parallel brain” to take a decision and pop up a message to the control system that “Go and check raw mill exit thermocouple, it looks fishy! “We can add some humor too.
- Anomaly detection and provide a real time alert to the control room:
Any Chemical process is dynamic we all know that. Every minute or hour it can change for myriad of reasons. SCADA or Plant Control System cannot figure out the anomaly. The control system is “dumb” until to the point when the process variable reaches either to the alarm or shutdown limit. There is a room between the operating range and the alarm and shutdown limit for critical variables. If a critical parameter is slowly going out of the norm, anomaly detection is the only means to track it and report it in timely manner. I do envision that Industry will have to have anomaly detection integrated with the control system, if not built in into it, at least an on-site edge computing should work in real time by pulling the Process data in every seconds, analyze it for presence of any anomaly and report it if the anomaly conditions and triggers are met.
Trust one day we will reach to a point where a “surprise” kiln downtime can be considered as “things of the past”. That being said, to get into this point we all need to do a tremendous work, we being everyone, the owners, operators and solution providers.
Manirul Haque/Houston/Texas.