Improving plant performance with intelligence and AI

Preparing for AI Controls (Field Devices and Data)

Like most people, I have begun using various AI in some parts of my work. It saves time and simplifies work in many ways. However - if I don’t supply adequate information, or if I fail to clearly define the objectives, the results are sometimes weak. It’s not the AI’s fault really – I just don’t always give it all the info it needs to provide the right solutions. 

On a parallel thought, lately I have been seeing lots of interesting articles on how AI will revolutionize industry, and I believe much of that is true. But based on my experiences, I foresee some problems which may hamper that progress that nobody seems to be talking about... So here goes.

To succeed, the AI will need way more data than most plants currently track/store, and it will need the final control elements to do what they are told to do…Most of the plants I have visited over the last 30 years still aren’t even properly implementing some of the technology that was “new” in the 90’s (or earlier)… And yet, some of those same organizations are actively considering implementing AI into their current plant.

Before implementing AI or other intelligence-based solutions, those plants should get their instrumentation measurements and their control devices up to full current capabilities. Here are some examples of areas that many plants are failing to capitalize on existing capabilities within the I&C field:

• SMART transmitters are often being implemented as dumb 4-20mA devices. Plants are still using maintenance assumptions based on long-gone technologies with very few (if any) of the powerful diagnostic capabilities being configured, actively monitored, or even looked at or recorded during maintenance.

• Intelligent valve positioners are often configured and treated as default (non-intelligent) devices with none of the powerful diagnostic capabilities being monitored, implemented, or even looked at, and the powerful failure options are rarely configured or utilized. In short, DVC’s are treated like really expensive I/P’s.

• Control valve position signals (ZT’s) are often not fed back to control system and/or to the historian system (and even if they are, they’re sampled at such a low sample rates that the data is useless for process control optimization work). Valve position feedback signals are one of the most valuable and informative pieces of information available to a process controls professional when analyzing control performance (and/or to a machine or AI system). Never skimp on sending the ZT’s to the controller...

• Important process variables are often not being monitored, and/or they are sampled or historized at grossly adequate sample rates.

For plants hoping to make use of data-centric technologies like AI, factor in the cost of the data points on historian licenses, along with the servers and other hardware needed to store and backup all that info – or don’t make the investment. Skimping on data collection and storage is not an option if you are moving to intelligent / AI systems. Without high resolution data the AI capabilities will be extremely limited.

As an example, valve positioners often cycle very quickly (several times per second of faster) – so to avoid data aliasing problems (confusion of the data), the sample rates may need to be in the 100 millisecond range (or even faster). Getting those data rates may be a challenge for some control systems due to bandwidth issues or limitations of the analog cards. And those data points will need to be available for a long time in order to do helpful long-term comparisons – meaning you can’t just revert to a 5-minute ‘average’ and dump the extra data after 24 or 72hrs.

• For context: A single analog data point at 100msec will eat up 320MB of data over the course of one year. Multiply that by 2000 to 100,000 data points and you get the idea… And in reality that data will be more than one byte due to floating point data, time stamps, etc.). 

• Being able to compare high resolution data (such as control valve position) to output commands and other process variables over time can provide incredibly powerful diagnostics, as well as process tuning & optimization insights.

These are just a few examples - but hopefully they give you an idea of what might be needed to get the plant running and ready for some of the super intelligent solutions coming down the pike. Still – it is incredibly exciting to think of what an AI system could do with the right data (and with the right setup, objectives, constraints, and guidance).

My bet is that a well-designed AI system could win the Indy 500 (car race) against the best drivers in the world, BUT ONLY IF THE CAR IS MECHANICALLY SOUND AND IF THE AI HAS ALL OF THE RELEVANT DATA.

Before sending tons of ‘whatever data you have’ to an intelligent or AI system and naively assuming it magically help your plant ‘win the Indy 500’ of your particular industry – you need to ensure that you provide it with the necessary data, and that the engine is tuned, the steering is tight, the brakes are working properly, and the tires are aligned and in good shape.

The Big Ugly (Control Valves)
Saying all of the above – from my experience, the single biggest cause of most process control problems is CONTROL VALVES and associated actuators, positioners, and air systems - since they are the moving parts that wear out and that get the most abuse. Yeah – I’m talking about those big, ugly, corroded, heavy metal monsters with grease and grime all over them.

They aren’t very sexy or cool compared to controllers, AI, or advanced control algorithms – BUT if those big ugly valves are neglected, nothing else is going to work right. If you don’t do the little adjustments and performance tweaks on the control valves, good luck getting the overall system perfected. 

Because of the whole ‘moving parts’ thing, and because control valves are often subjected to the most brutal conditions imaginable, problems are going to arise, such as; Cavitation and flashing will cause wear and will change the valve characteristics over time, stems will corrode and become sticky, linkages and tension springs get looser over time and will eventually result in position sensor errors, position potentiometers and rollers get bumps, noisy spots or flat-spots, and the list goes on. Talk to a knowledgeable and experienced instrument tech or control valve specialist for a few minutes and they’ll quickly add dozens of items to this list…

So – First, make sure the control valves are fully functional before you go too crazy on the controller or other aspects. If you get the control valves (and other final control elements) working correctly, AI and the plant controllers and everything else has a chance.

If the final control elements are not confirmed to be working 100% correctly, the controls side of the loop is very limited and simply won’t be able to achieve top levels of performance, regardless of how genius the algorithms, strategies, or intelligence of the control system or the operator.

As an example: Pretend the tension spring for a valve positioner sensor has degraded or broken resulting in 3% hysteresis between upscale and downscale readings. Even the best, most attentive operator in the world, will NOT be able to maintain the PV at setpoint with that system. It IS going to fluctuate, period. AI won’t be able to fix that problem – so it will do what operators do; it will find the best ‘available’ solution that it can come up with based on what it can see and what it can control. But the real (and only good) solution would be to identify and repair the positioner tension spring. For AI to ‘see’ this problem it would need data that it likely would not be getting (such as high baud data on valve ZT and/or the digital diagnostics HART / Fieldbus data from the positioner). 

Many times, the valve positioners aren’t even configured to notice, or to report deviations, changes, alerts, or other potential problems – and sometimes, even if they are configured to notice and alert, the alert just sits there inside the positioner and nobody knows unless they happen to connect a communicator to it and look for the alert.

So that alert that could curtail a trip or a safety incident just sits there inside the positioner's brain, as the problem gets worse and worse until it causes a hard failure or plant shutdown.

It doesn’t make any sense at all to pay for an expensive, super smart device like a DVC that has an “I knew this would happen” bit, and to tell it to ‘sit down and shut up’ – but that is exactly what happens most of the time.

Question to Gurus – Other than full-blown solutions like THUM’s, wireless, or HART analog cards, do any of the valve positioner manufacturers produce any kind of super low cost, local visual/audible annunciator (a blinky or a buzzy) to help alert operators to a diagnostic alert/warning within an intelligent positioner? I don’t recall any off the top of my head, but it’d sure be helpful for those plants that don’t have any way for the operators to know a valve positioner has an alert to report to us humans.

Nearly all modern smart (or digital) positioners have incredible diagnostic capabilities and have the ability to report developing problems long before they cost money or cause potential safety issues.
Over the years, I’ve talked with various vendor reps about this issue, and they admit the frustration of seeing their incredibly powerful equipment utilized at a fraction of it’s full capability in so many plants.

This is true for measurement, controls, electrical, and even mechanical systems. In short - The equipment you already own is likely capable of way more than what it is currently doing. Often times it could do it with little or no additional investment, other than having a knowledgeable person actually do the research and properly configure some things…

So – To prepare for possibly investing millions into new automation technologies, start by doing the following (5) things: Each of these will help the plant NOW, and each will be essential for any real success with any intelligent or AI solutions.

1) Get training - Get your techs and engineers well trained on the current technologies like HART, Smart Instrumentation, Digital Valve Positioners, and other IO devices you may use such as; VFD’s, Soft-Starts, Intelligent MCC’s, etc..

a. Educate them on: How the systems work (in detail); How they differ from older technologies; How they impact maintenance plans and procedures; How the technologies can be implemented to enhance reliability, and How the system/technology is implemented at their specific site. If the folks doing the configs and replacing the parts don’t understand the technologies and equipment, things aren’t going to progress well.

b. Example - Some of the ‘new’ technologies that many techs need training on are things like HART/SMART vs analog instruments. This technology is not really ‘new – but a huge portion of techs and engineers across all industries are very unfamiliar with the HART/FieldBus and SMART operation or capabilities. Many senior techs and engineers assume a Smart transmitter is just a ‘more accurate’ version of an older analog/non-smart instrument and don’t really understand the differences.

2) Get technical help – Assign a knowledgeable person to study your I&C equipment and systems and lay out the issues and possibly help formulate a plan to capitalize on existing technologies and the realities of your plant. If you don’t have anyone with appropriate knowledge, experience, and credentials, hire an accredited and proven 3rd party consultant.

3) Establish the data feeds - Set up to collect (and historize) the relevant data points at the necessary sampling rates. Yes, this will cost money – but without that data, you are blind to see or study the problems and you have little hope of finding optimal solutions (and AI would be blind as well).

4) Fully capitalize on the current capabilities of equipment you already own - Strategically and systematically setup and configure your existing I&C equipment to capitalize on it’s full existing capabilities:

a. Setup to capture direct valve position feedback via a fast signal (typically analog with at least 100msec data capability).

b. Setup to pull positioner diagnostic data & alerts into the control system. This can be pulled in digitally via HART, FieldBus, or other methods. Instrument vendors will be happy to discuss the options to get this data into your control system. There are so many advantages to having this powerful data delivered to the control system. Operators can be notified of even tiny changes in valve performance and can be proactive (vs reactive) to solving issues before they cause trips and cost money.

There are many good ways to gather this diagnostic data without interfering with or altering your current control system IO racks and since it is ‘monitor only’ data it isn’t subject to some of the same requirements as other data. In short – it’s probably easier than most would assume to get this data into the system.

c. Create a template form to establish some strategic and consistent configurations for your valve positioners. This should include failure modes & options, deviation alerts, and so on. This will require some thought (and possibly some digging in the reference manuals or input from vendor experts) and may vary by system & situation. Plan on updating your instrument Spec Sheets as part of this process (aka Config Sheets) since they are likely don’t include all the relevant data fields and settings that need to be recorded and maintained.

For example: If you don’t have a record for how instruments (including valve positioners) are to be configured, tuned, and setup, there will be no consistency, and it will inject many unknowns and wildcard variables into the equation of your plant controls.

Did you know that most modern intelligent valve positioners will notice and send alerts when things change in valve operation such as changed supply pressure, increased stiction due to someone over-tightening the packing, linkage slop, broken actuator spring, etc. Many (but not all) problems can be reported and then proactively resolved before causing any problems. Many can even revert to I/P mode if the ZT sensor fails. Why are so few organizations taking advantage of these incredible features???

d. Same topic as above - but for your process transmitters. Very few of the powerful and helpful diagnostic capabilities of today’s instrumentation are being utilized in the vast majority of plants.

Did you know that those ‘little blue’ pressure transmitters can actually give you an alert when moisture in a j-box starts altering the 4-20mA signal and causing inaccuracies (see my other blog on this issue for more) or that they can recognize when an impulse line is clogged. Why would we not take advantage of these powerful diagnostic alerting capabilities? Only about 10% of the sites I have personally visited are taking advantage of ANY of those types of diagnostics on ANY of their capable instruments – yet MOST of their instruments have these types of diagnostic capabilities.

5) Test key control valves! - Periodically perform a full TEST on your control valves (especially on critical, problematic, or very fast processes) and then FIX the problems you find. If your final control elements are not in good working order, no amount of algorithms or genius controls will be able to produce good results *See the note at end of this article for tips on how to conduct truly effective and complete valve testing.

Doing the 5 things above will drastically improve your plant performance AND will help get you ready to capitalize on some of the awesome new technologies that are coming fast.

Summary
The best race car driver in the world – (or the most powerful supercomputer / AI system), would not be able to win the Indy 500 race in a car that has a sluggish engine or a sticky throttle actuator, or that has slop in the steering, or with tires that are imbalanced, out of alignment, or have poor tread.

“Get the car running – then improve the driver”

*Note – For testing control valves, I would suggest a system like the Profiler™ system by SofTek Engineering (Softekeng.com). They provide a portable, easy to use system that tests for nearly every type of problem that can occur with about any industrial valve / actuator / positioner combo (literally including air supply, positioner, actuator, and the valve body itself). It’s a software-based system that is extremely well automated (including the reports and analysis). Crazy cool stuff.
In my opinion, being able to perform a full test of any control valve is a requirement for any organization hoping to achieve any strong level of process control.

My suggestion would be to test all control valves (and establishing baseline data) when you first install them to help diagnose and solve the various problems that show up over the life of the process. In addition, I'd suggest testing key valves at various intervals (such as TARs), depending on criticality and based on failure and problems findings with various brands and models used on various applications at your facility.