Data Corruption - Part 4 (Physical Output Devices)

More causes of bad data...

This series on Data Corruption in Control Systems explores the most common—yet frequently overlooked—problems affecting industrial control system performance and associated data & analytics.

In part 1, we discussed Signal Filtering. In part 2 we covered Digital Resolution & Degradation issues. In part 3 we covered Input Accuracy & Precision issues. In part 4, we will cover Output Device Behaviors & Impacts. In part 5, we will cover the issues caused by the controller and the process itself. 

Data Corruption - Part 4 (Physical Output Device Problems)

While considerable attention is often given to transmitter accuracy and controller tuning, the mechanical output devices--valves, actuators, and positioners--frequently don't receive the same level of scrutiny.

Yet, these can be major sources of control problems and data corruption that propagate throughout the entire system.

Outputs (like controllers and process issues we will cover in the next part) produce errors indirectly – by impacting the process, which propagates through to the measurements. This is not as intuitive of a path to data errors and therefore gets less attention than it should.

Outputs such as control valves also cause many of a plants controls issues and problems, for many of the same reasons that they can cause data problems. In fact, the control problems are usually the root source of the data problems caused by output elements.

Before we begin, let’s analyze a common scenario to help explain how a physical field output problem can manifest into not only a controls problem, but can also degrade data accuracy and quality.

Example Scenario: Valve Packing Over-tightened

Consider this common scenario:

1. An operator decides to tighten the packing of a control valve to address a minor leak and inadvertently adjusts it too tightly.
2. The valve packing now imparts excessive friction on the actuator stem.
3. The high friction results in a sticky behavior of the actuator (friction/stickiness--commonly referred to as "stiction").
4. The stiction causes the valve to only respond when the pressure applied to the actuator can overcome the friction forces, resulting in jerky (non-smooth) movement of the stem.
5. Because of this jerky movement, the positioner struggles to position the valve accurately, causing small errors in PV. Even if the process was inherently stable, as the positioner continues to seek the desired valve position setpoint, it will fluctuate above and below the target--producing real variations in the process that ultimately affect the PV values and show up as a form of induced (secondary) noise.
6. In effect, the process is now perpetually cycling (in a somewhat random way) due to positioner attempting to find the proper position as it chases the valve command signal, which is being modulated due to the variations.
7. There are many factors that will affect the shape, amplitude, frequency, and degree of randomness of the fluctuations of the valve position and of the process variable being measured.  

Note on Process Simulations

In addition to actual bad data stemming from outputs, another important issue is how outputs are treated in SIMULATIONS:

• If a simulation does not include the behaviors that happen "after the controller"--such as the real-world deadtimes, lags, gains, stroke speeds, capacities, integral characteristics, stiction, deadband, hysteresis, and other issues discussed in this article--the simulation will simply NOT be accurate.

• In the real world, there are just as many complexities in the valve positioner, actuator, and valve as there are in most processes--so they should be factored in.

• Many models treat valves as ideal entities with at best a valve characteristic and a Cv number entered--and of course, they don't respond like the real processes, especially on faster loops with larger valves (such as recycle flow).

• While we won't go deeper into simulations in this article, any process simulation should recognize the complex behavior "downstream" of the controller and factor it in if the simulation is to be used for anything beyond general responses or operator interface or procedures type training. Many OTS (operator training systems) are poorly suited to any serious engineering / process control work because they fail to factor in the real-world factors that fall outside the internal process physics or the controller algorithms (such as valves, actuators, positioners, etc.).

• The worst-case scenario is where a plant uses a model or simulation that doesn't account for the realities of output devices, AND also fails to address the problems that exist in those devices -- so the errors are in the real world, but not in the simulation.

Why Mechanical Output Problems Corrupt Data

Mechanical output issues don't just affect control performance—by altering the control functionality they directly alter and corrupt the data that's being collected from your process by injecting secondary noise fluctuations or oscillations and/or other errors into the actual process:

Oscillations Propagate

When positioner issues create oscillations, they don't remain isolated to the valve itself but propagate throughout the entire process, creating data corruption far from the source:

• Amplification Through Process Dynamics: Small oscillations in valve position can be amplified by process dynamics, particularly in systems with high gain or resonant characteristics. For example, a barely perceptible 0.5% valve oscillation in a reactor feed valve can create significant temperature oscillations if it excites the natural frequency of the reactor.

• Cross-Variable Contamination: Valve oscillations in one process stream contaminate data in seemingly unrelated variables. For instance, minor hunting in a cooling water valve can create oscillations that appear in product quality measurements, leading analysts to incorrectly focus on product formulation rather than mechanical issues.

• Masked by Process Noise: The propagated oscillations often blend with normal process variability, making them difficult to distinguish from legitimate process dynamics. Frequency analysis may be necessary to separate these mechanical signatures from actual process behavior.

• Cascading Effects: In interconnected processes, oscillations from one control loop can cascade through multiple units. A single problematic valve can create data corruption across an entire process train, with the signature becoming more distorted (but still present) as it propagates.

• Time-Delayed Appearances: Due to process deadtime, the oscillations may appear in different measurements with varying time delays, creating the false impression of complex, time-dependent process interactions.

As Michael Bauer noted in his Control Engineering article on valve diagnostics: "The true impact of a cycling valve extends far beyond its immediate loop. The oscillations become embedded in the process itself, creating a data signature that persists downstream and can trigger false alarms in other units. Many 'mysterious' plant-wide oscillations can be traced to a single mechanical issue when properly analyzed." [11]

Response Time Problems

The speed mismatch between mechanical response and process dynamics creates significant data corruption that undermines control performance and analytics:

• Minimum Control Performance Limitation: Just as the sampling rate must be significantly faster than process dynamics in digital control, the mechanical response must be substantially faster than the process to achieve good control. When valve response time approaches 20% of the process time constant, control performance degrades markedly and data quality suffers.

• The Cascade Control Reality: Every valve with a positioner is actually operating as a cascade control system--the positioner is the secondary (inner) loop controlling valve position, while the process controller is the primary (outer) loop. This cascade relationship means all the classic cascade control timing requirements apply: the inner loop must be significantly faster (typically 3-5 times) than the outer loop.

• Valve-Process Speed Mismatches: When valve response is too slow relative to the process:

* Control loops become dynamically limited by the valve, not the process
* Controllers must be detuned to prevent oscillations, increasing overall variability
* Process data shows characteristic "sluggish then overshoot" patterns that don't reflect true process dynamics
* Fast disturbances pass through the system uncontrolled, creating data points that appear as outliers 

• Impact on Faster Control Loops: This is particularly problematic for flow, pressure, and some composition loops where process dynamics are naturally fast. A valve that performs adequately for slow temperature control may be completely inadequate for fast flow control, creating inconsistent data quality across different process variables.

Gregory K. McMillan explains in "Control Loop Foundations": "When valve response time approaches the process time constant, the effective loop deadtime doubles and the maximum achievable control performance drops by 50% or more. The resulting data shows a characteristic rounded response pattern that's often misinterpreted as process nonlinearity when it's simply the valve unable to keep pace with the process dynamics." [12]

False Process Dynamics

One of the most insidious ways mechanical output problems corrupt data is by creating patterns that mimic legitimate process behaviors:

• Cyclical Patterns: Stiction in a valve can create a characteristic saw-tooth pattern in the process variable that resembles a periodic disturbance or cycling feed composition. Data scientists may spend considerable time looking for a process cause when the issue is purely mechanical.

• Apparent Time Constants: Sluggish valve response due to oversized actuators or positioner issues can make the process appear to have longer time constants than it actually does. This leads to models that overestimate process inertia and underestimate the potential control performance.

• False Process Interactions: When valve problems affect multiple streams simultaneously (e.g., in a heat exchanger with interacting temperature and flow control), the resulting patterns can appear as process interactions. Analysts might conclude that process variables are physically coupled when the relationship is actually created by the mechanical issues.

• Misleading Process Nonlinearities: Valve deadband creates a characteristic "wait-and-jump" response that makes the process appear nonlinear or to have variable gain. This can lead to unnecessarily complex models that try to capture this "process behavior" when it's actually a mechanical artifact.

• Artificial Disturbance Magnification: In some cases, mechanical issues selectively amplify certain process disturbances while dampening others, creating a distorted picture of which disturbances truly impact the process most significantly.

The practical consequence is that engineers and data scientists may develop sophisticated models and control strategies to address these "process dynamics" that don't actually exist in the process itself. This diverts resources from addressing the root mechanical issues and can lead to complex solutions for problems that could be solved through proper valve maintenance or configuration.

As noted by Bialkowski in his classic paper on control valve performance: "Process models derived from plant tests often incorporate the valve dynamics unintentionally, leading to compensatory control strategies that attempt to overcome mechanical limitations rather than optimize the actual process." [10]

Compounding Effects

Mechanical output issues can also compound with input measurement problems discussed in previous parts, creating layered corruption that's difficult or impossible to analyze accurately.

These patterns of data corruption often lead to misdiagnosis of process problems, incorrect tuning adjustments, and faulty analytics conclusions.

Ok -- so we covered how mechanical issues can cause data errors - Now, lets cover the sources of some of these mechanical problems.

 

1. Valve Positioner Issues

1.1 Positioner Tuning Problems

Positioner tuning is often overlooked as a source of data corruption. While tuning is typically as simple as entering valve/actuator information in the fields of a Digital Valve Controller (DVC), there are factors that complicate this:

• Detuned Positioners: Sometimes, technicians intentionally "detune" the positioner in an attempt to overcome or mask other problems, such as stiction or booster mis-adjustments or other problems. If the positioner is detuned, it will lead to delays in reaching setpoint which (if slow enough) can result in interactions with the process loop and potential oscillations at certain conditions.

• Aggressive Tuning: Conversely, if the positioner is tuned too aggressively, it may constantly cycle and hunt for a precise target, resulting in small secondary process noise errors that may confuse intelligent systems analyzing the data.

• Valve Position Hunting: On systems that have valve position fed back to control system and/or analytical systems, comparing the valve position to command signals can provide tremendous insights to the overall valve command system.

• Constant fluctuations in the valve position independent of the valve command can help identify problems with various parts of the command-positioner-actuator-valve system. 
• Further comparing this signal to the valve flow or other related PV as well as to other associated or potentially interacting systems is fantastically useful data. Many times this data is excessively filtered to reach it's full potential.

As noted by Michael Taube in Control Engineering: "Positioner hunting creates a characteristic high-frequency component in process data that's commonly mistaken for electrical noise. This can lead to excessive filtering that masks actual process dynamics, creating a compounding problem for both control and analytics." [1]

1.2 Position Sensor Feedback Problems

Position sensor issues are particularly common with older positioners:

• Linkage Problems: This is a common problem with older position sensors (prior to the Hall Effect-type sensors now common on many DVCs). Any linkage in the sensor system can and will eventually cause problems due to wear, loosening, or improper adjustments.

• Potentiometer Issues: Older style systems used rotary potentiometer sensors that often developed "glitch spots" in their position curve and also sometimes produced noise and spikes while rotating. The ZT signal from these systems could cause confusion for the positioner and/or for any controller logic or higher level data analysis.

The best solution for older-style sensors is replacement or upgrade. They are a significant problem for both control and data quality.

All serious control systems should include a ZT feedback from any important control valves (and/or critical isolation or shutdown valves) as a core component. This single data point is one of the most powerful tools available to a strong controls engineer when analyzing plant performance problems.

1.3 Air System Issues

Air systems are often neglected and overlooked sources of problems on outputs:

• Supply Problems: Inadequate supply pressure, fluctuating header pressure, or undersized supply lines create inconsistent actuator responses.

• Air Quality Issues: Few plants perform the required maintenance on dryers, local regulators, and moisture separators. Poor air quality leads to internal corrosion, sticky operation, and eventual failure.

• Contamination: Few plants have adequate controls on keeping the systems sealed during maintenance or preventing contaminants such as Teflon tape, dust, or other particles from getting inside the system.

• Note on Air Systems: If a plant doesn't treat their instrument air just as critically as they treat fire & gas systems, or SIS systems, they aren't seeing the big picture. Instrument Air is one of the primary root causes of failure for many/most I&C failures.

• For sake of this article, badly designed or poorly maintained air systems are going to cause those tiny orifices in the I/P's in valve positioners to get partially (or completely) clogged, resulting in erroneous I/P behavior and subsequent positioner problems which lead to the errors in the PV.
  
  
• It is all connected… and instrument air feeds nearly every I&C valve. Improving instrument air systems and taking the associated maintenance seriously is low hanging fruit for plant improvements.

According to the ISA Standards & Practices Department's "Quality Standard for Instrument Air" (ANSI/ISA-7.0.01): "Inadequate instrument air quality is responsible for approximately 25% of emergency shutdowns in continuous process industries, many of which can be traced to control valve failures that would have been visible in data trends prior to the incident." [2]

1.4 Control Valve Volume/Pressure Booster Problems

Boosters are critical yet frequently misunderstood components:

How Boosters Work

Boosters are installed between a positioner and actuator to increase the air flow rate to large actuators, allowing faster valve movement than the positioner alone could provide. They maintain the same pressure as the positioner output but deliver it at a much higher flow rate.

Common Booster Problems

• Interdependent Adjustments: Boosters and positioner tuning must be coordinated very carefully. The booster's bypass adjustment significantly impacts the overall dynamic response of the valve system
• Balancing Speed vs. Stability: A booster that's improperly adjusted creates a dilemma:
  
  • Too little bypass: Increases valve speed but introduces instability and oscillations
  
  • Too much bypass: Provides stability but defeats the booster's purpose by slowing response
  
  • Data Corruption Pattern: The signature of booster problems in data is typically a slow initial movement followed by a rapid overcorrection as the booster engages, creating a characteristic "S" curve in valve position data. This pattern occurs because:

1. 1. The initial valve movement is delayed while pressure builds
   2. Once the booster activates, it delivers air much faster than needed
   3. The valve overshoots the target position
   4. The positioner tries to correct, creating oscillation

Proper Adjustment Sequence

According to Emerson's "Control Valve Handbook" and Fisher technical publications on booster applications, the proper sequence for adjusting boosters and tuning positioners is:

1. Start with Maximum Bypass: Begin with the booster bypass fully open
2. Tune the Positioner First: With maximum bypass, the booster has minimal effect. Tune the positioner to stable operation.
3. Gradually Close the Bypass: Incrementally close the bypass valve while monitoring valve response.
4. Make small adjustments (¼ turn)
5. Test with step changes after each adjustmen
6. Look for smooth, responsive movement without oscillation
7. Find the Stability Threshold: Continue until you observe the first signs of instability or oscillation, then back off slightly.
8. Re-verify Positioner Tuning: The system may need positioner retuning after booster adjustment.
9. Document the Configuration: Record the final settings for future reference

Impact on Data Quality

As noted by Fisher Controls in their technical paper "Volume Booster Performance in Digital Valve Applications": "The dynamic response of a valve with an improperly adjusted booster creates characteristic patterns in process data that mimic process deadtime followed by overreaction. This signature is often misinterpreted as process dynamics when it's purely a mechanical artifact." [13]

The Valve Accessories & Controls technical guide adds: "Data collected from valves with improperly adjusted boosters shows artificial deadtime followed by rapid jumps in position. This creates what appears to be process nonlinearity in the data, when it's actually the booster's dynamic response characteristic." [14]

Detecting Booster Problems

Gregory K. McMillan, in his "Essentials of Modern Measurements and Final Elements," provides this guidance for detecting booster problems in process data: "Look for a characteristic pattern where small setpoint changes create almost no response for several seconds, followed by a rapid jump past the target position and subsequent oscillation. This pattern occurs at a consistent frequency regardless of process conditions--a telltale sign it's a mechanical rather than process issue." [15]

Locking Down Configurations

As Fisher's "Volume Booster Application Guide" recommends: "Once properly configured, booster bypass adjustments should be mechanically locked or sealed to prevent unauthorized adjustments. A single well-intentioned but improper adjustment can destabilize the entire control loop, creating persistent oscillations that are difficult to diagnose." [16]

1.5 I/P Transducer Problems

I/P (current-to-pneumatic) transducer issues are frequently overlooked but critically important:

• Orifice Contamination: Even the latest digital positioners still include a tiny I/P transducer inside with very small orifices that can easily become clogged. Any moisture or contaminants in the system lead to poor I/P response.

• Partial Blockage: Tiny bits of rust, Teflon tape, dust, or other contaminants will seek out the small orifice. Sometimes they clog it entirely, but often they block it just enough to throw off the transfer curve or make it nonlinear.

• Diagnostic Importance: A well-built Industry 4.0 plant should already be monitoring the diagnostic data of positioners. Getting diagnostic data and alerts from positioners to HMI/operations should be a top priority of any controls advancement project.

Mike says; "If you implement only one diagnostic capability in your entire plant, it should be positioner diagnostics - knowing when your control valves aren't working properly is a a sign that you are about to be having problems and losing money or worse..."

The ISA Process Automation Committee notes in its technical report on Smart Device Integration: "Modern digital positioners can detect I/P issues before they become severe enough to affect process control. Plants that integrate this diagnostic data into their condition monitoring systems see a 30-40% reduction in control valve related process disruptions." [3]

1.6 Positioner Configuration Mistakes

Configuration errors are common with modern digital valve controllers:

• Excessive Deadband: Often configured to overcome other mechanical problems, excessive deadband creates zones where small process changes receive no control response.

• Incorrect Characterization: Mismatched valve characterization curves (linear, equal percentage, etc.) create inconsistent control responses across the operating range.

• Cut-off Values: Improperly set cut-off values create artificial discontinuities in control response.

• RTFM Principle: Most of these issues are specific to the type of positioner used and are typically well explained in the manuals, but few engineers and fewer technicians ever read them.

The RTFM acronym (Read The Functional Manual) applies to any complex device such as positioners. Attempting to hack one's way through a DVC setup will often lead to errors and mistakes like those above and result in bad control and bad data. If it's anything out of the ordinary or if you don't understand a setting, parameter, or behavior, pause and RTFM.

2. Mechanical Valve Problems

2.1 Deadband and Hysteresis

Hysteresis and deadband represent perhaps the largest category of problems affecting data quality in process plants:

Deadband Defined

Deadband is the gap between when a command signal changes and when the valve actually moves. This creates a "dead zone" where process variables drift uncontrolled, introducing apparent randomness in data that isn't actually process-related.

Primary Causes of Deadband:

Mechanical Friction: Static friction (stiction) between valve stem and packing creates resistance that must be overcome before movement occurs. 

• Backlash in Linkages: Play or looseness in mechanical connections between actuator and valve trim creates lost motion. It is surprising how much backlash there is even on directly 'hard-coupled' or bolted linkages such as on rotating or linear stems.

• Worn Valve Components: Worn stems, bushings, or guides create increased clearances on actuators and mechanical components that must be traversed before effective movement occurs. Even with tension springs there will be some amount of deadband and hysteresis in any mechanical system.

• Positioner Deadzone: Some positioners are designed with intentional deadzone to prevent constant micro-adjustments that would cause premature wear. These deadzones will result in a repeating pattern of slowing drifting and then a sudden correction but the size and timing of the cycles may vary based on many factors and is likely to be misinterpreted in analysis.

Hysteresis Defined

Hysteresis refers to the difference in valve  position between increasing and decreasing signals at the same command value. This means your process is actually following two different control curves--creating what appears to be inconsistent data when in fact it's a mechanical problem.

Primary Causes of Hysteresis:

• Elastic Deformation: Components briefly deform under load then recover differently when unloaded. 
  
  
• Friction Directional Differences: Moving parts experience different friction forces depending on direction of travel.
• Actuator Spring Characteristics: Compressed springs exhibit different force-distance relationships during compression versus relaxation.
  
  
• Backlash in Connection Points: Play in connection points causes movement differences when changing direction.
  
  
• Air System Trapped Volume: Trapped air in actuator systems compresses differently during pressurization versus depressurization

Data Corruption Impact

Deadband creates a "stair-step" pattern in process data where the process variable doesn't change until deadband is overcome, then jumps suddenly. This pattern is often mistaken for process instability or noise when it's actually a mechanical issue.

Hysteresis causes a difference in the upscale and downscale results. A small hysteresis error on a positioner is easily corrected via the typical PID control output simply modulating the command signal, but a larger amount of hysteresis could result in the controller repeatedly hunting or bumping slightly as small errors build since the system will not respond in direction opposite of what it last developed until the error is greater than the hysteresis.

Process Implications:

• Process variables drift within the deadband/hysteresis range.
  
  
• Control loops become unstable as controllers increase output to overcome deadband.
  
  
• Loop oscillation frequency directly correlates to deadband magnitude.
  
  
• Standard tuning methods fail as the effective process gain changes with direction.
  
  
• Data analysis shows inexplicable "bands" of operation rather than consistent behavior

Recognizing Hysteresis and Deadband

Look for inconsistent valve position at the same controller output, "stair-step" position feedback response to smooth control signals, and different PV behavior with increasing versus decreasing signals.

According to Lipták in Process Control and Optimization: "Valve deadband is often the limiting factor in achievable control performance. A control valve with 2% deadband cannot realistically be expected to provide control performance better than ±1% under the best circumstances." [4]

In a Control Engineering article, Michael Taube notes: "Data patterns caused by valve hysteresis can fool analytics algorithms into seeing oscillatory behavior when what actually exists is simply a mechanical limitation. Predictive maintenance algorithms, in particular, can generate false positives when these patterns aren't properly recognized as mechanical signatures rather than process degradation." [5]

2.2 Actuator Design Issues

2.2.1 Incorrect Spring or Bench Set

Spring and bench set issues directly impact valve position accuracy:

• Spring Range Mismatched: When the actuator spring range doesn't match the application pressure requirements, the valve operates in a non-optimal portion of its travel range.

• Incorrect Bench Set: Improper bench set causes the valve to be open when it should be closed (or vice versa) at specific signal values, creating positional errors that directly corrupt data.

• Asymmetric Response: An incorrectly configured spring range creates asymmetric valve responses--faster in one direction than the other--which introduces directional biases into the process data.

According to McMillan in Good Tuning: A Pocket Guide: "Improper bench set is often responsible for unexplained loop oscillations, particularly in tight shutoff applications. The valve may be fully closed at 4mA, but begin to open far above the expected point, creating an effective deadband that destabilizes the loop." [6]

The ISA Control Valve Handbook notes: "Data collected from a process with incorrect actuator configuration will show characteristic asymmetric responses to setpoint changes--faster in one direction than the other. This pattern is often misinterpreted as process nonlinearity when it's actually a mechanical configuration issue." [7]

2.2.2 Inadequate Actuator Stiffness

Actuator stiffness problems create inconsistent valve responses that corrupt process data in subtle ways:

Force Balance Fundamentals

• Spring Selection Principles: Control valve actuator springs are specified by two key parameters:

• Spring Rate: Measured in force per unit distance (lbf/in or N/mm), this defines the spring's stiffness
  
  
• Bench Set Range: The span of input pressure that moves the valve from 0% to 100% position when no process forces are present. 
• Force Balance Issues: A valve that can barely move against its spring will struggle to achieve consistent positioning. Worse yet, a valve system that must handle high differential pressures (e.g., 1000 psi across a 1" valve plug) needs substantial actuator force to provide symmetrical and consistent response at all operating pressures.
  
  
• Spring Selection Impact: Springs must be selected to provide:
  
  
• Sufficient force to overcome maximum process forces (pressure drop × plug area)
  
  
• Consistent performance across the entire valve travel
  
  
• Appropriate response speed without sacrificing stability

Stiffness Factor Optimization

The "stiffness factor" is a critical specification that determines a valve's ability to maintain position: 

• Stiffness Factor Defined: The ratio of actuator force to the maximum process force that could be exerted on the valve trim
  
  
• Ideal Range: Industry standards suggest a minimum stiffness factor of 2.5-3.0 for consistent control
  
  
• Critical Applications: High-performance control applications may require stiffness factors of 5.0 or higher.
  
  
• Methods to Increase Stiffness Factor:

1. 1. Larger Actuator Diaphragm/Piston: Increasing effective area creates more force for the same pressure
   2. Higher Supply Pressure: Using higher air pressure (e.g., 60 psi instead of 35 psi) increases available force
   3. Multi-Spring Actuators: Combining springs in parallel increases force without excessive compression
   4. Stiffer Spring Materials: Higher spring rates maintain position better against varying process forces
   5. Actuator Boosters: Pneumatic boosters can increase effective force without changing actuator size
   6. Double-Acting Actuators: Applying air to both sides of the diaphragm/piston and controlling the differential
   7. Smart Positioners with Adaptive Control: Digital positioners that compensate for varying process forces

Process-Induced Asymmetry

When process forces oppose actuator movement in one direction but assist in the other, the result is asymmetric response--the valve moves at different rates depending on direction:

• Unbalanced Plug Designs: Pressure on the upstream side of an unbalanced plug creates additional force proportional to the pressure drop

• Flow Direction Effects: Flow direction can create lift or drag on valve trim, adding or subtracting from actuator force

• Dynamic Process Forces: At high velocities, fluid dynamic forces can vary dramatically with trim position

• Position-Dependent Behavior: Process forces can change substantially across the valve's travel range, creating nonlinear responses

Data Impact

Stiffness problems create patterns in the data that suggest process nonlinearities when in fact they're mechanical in nature. This leads to faulty conclusions about process behavior:

• Apparent Process Gain Changes: When a valve struggles against process forces, it creates the appearance of varying process gain

• Pressure-Dependent Behavior: The process appears to behave differently at different system pressures

• Flow Rate Anomalies: Control loops show different behaviors at high versus low flow rates

Recognizing Stiffness Problems

Look for these telltale signs in process data:

• Slow response in one direction versus the other: Particularly noticeable when changing setpoints up versus down

• Inconsistent positioning at different process conditions despite identical control outputs: The same controller output produces different valve positions under different differential pressures

• Position hunting at specific operating points: The valve struggles to maintain position when process forces approach spring forces

• Correlation between system pressure and control performance: Control performance deteriorates at specific pressure ranges

• Stepped response to smooth signals: The valve moves in discrete jumps rather than smoothly following signal changes

As Fisher Controls notes in their valve selection guide: "Insufficient actuator stiffness creates position uncertainty proportional to process force variation. Each 1% of position uncertainty translates directly to process variability, creating apparent process dynamics that don't actually exist in the process itself."

The Control Engineering article "Addressing Actuator Stiffness in High-Pressure Applications" notes: "The data signature of inadequate actuator stiffness includes higher variability at high differential pressures and asymmetric response to setpoint changes. This creates a false impression of process nonlinearity that can lead to excessive controller detuning, further degrading overall performance." [8]

2.3 Valve Sizing Problems

Improper valve sizing creates multiple data corruption issues:

• Valves Operating Near Saturation: When control valves are operated at or near saturation they can produce a Reset Windup issue due to integral action not producing a change in PV. Once initiated this windup can result in an oscillation of the process due to the slow wind-down of the integral component after each windup period.

• Oversized Valves: Operate in the difficult-to-control bottom portion of their range, creating excessive sensitivity to small signal changes and instability.

• Undersized Valves: Cannot provide adequate flow and saturate at less than 100% output, creating artificial limits in process data.

• Choke Valve Interactions: Valve saturation behaviors can also be impacted by an inline flow control choke (flow control bias). In some systems, a choke valve in line with a control valve can be used to "bias" the overall pressure drop of the system and effectively reestablish the valve operating point so it better matches its Cv. If this valve is operated without oversight and/or without the analytical system knowledge, it could result in confusion or misleading data.

The ISA Standard "Control Valve Sizing Equations" (ANSI/ISA-75.01.01) states: "Operating a control valve below 20% or above 80% of its travel range significantly increases nonlinearity and reduces control capability, which manifests in data as unexplained variability at these operating ranges." [9]

2.4 Valve Characterization and Bias Interactions

Improper valve characteristic selection and valve bias configurations create systematic data corruption that can severely impact analytics:

2.4.1 Valve Characterization Fundamentals

• Linear vs. Equal Percentage Valves:

• Linear valves: Provide the same change in flow for a given change in valve position throughout the range
  
  
• Equal percentage (EP) valves: Provide flow changes that are proportional to the existing flow rate for a given change in valve position
• Quick-opening valves: Provide large flow changes with small position changes near closed position, then diminishing effect as valve opens further.

Matching Valve to Process:

• Processes with linear pressure-flow relationships (like many liquid flows) are best controlled with equal percentage valves
  
  
• Processes with exponential pressure-flow relationships (like many gas flows) are best controlled with linear valves
  
  
• Batch charging applications may benefit from quick-opening characteristics for rapid initial response

2.4.2 System Linearization Through Valve Selection

EP Valve Benefits:

• Counteracts the square-root relationship between pressure drop and flow
  
  
• Creates more consistent loop gain across operating ranges
  
  
• Particularly effective in systems with high pressure drop ratios (valve ΔP / system ΔP > 0.3)
  
  
• Essential for processes requiring wide turndown (>5:1 flow range)

Optimal Application Conditions: 

• EP valves work best when they have sufficient authority in the system
  
  
• Performance degrades when valve pressure drop is a small percentage of system drop
  
  
• Most effective when sized to operate in the 20-80% travel range.
  
  
• Should be paired with appropriate positioner dynamics for best results

Digital Characterization Options: 

• Modern digital valve controllers can implement custom characterization curves
  
  
• DCS systems can apply signal characterization before the valve
  
  
• Combination approaches can address both valve and measurement nonlinearities

2.4.3 Data Corruption Effects

• Mismatched valve characteristics create varying controller sensitivity (gain) across the operating range, resulting in fluctuations or oscillations at one end, and a sluggish response on the other.
  
  
• The resulting fluctuations or oscillations can produce misleading data points.

2.4.4 Analytics Impact

• Inconsistency creates operating point-dependent relationships between variables
   
• Correlation analyses show weak connections that are actually strong, but nonlinear
• Predictive models require nonlinear techniques to capture relationships
  
  
• Classification algorithms may create artificial categories based on valve operating regions
  
  
• Statistical process control limits may need to vary by operating range

Lipták in Process Control and Optimization states: "A common mistake is installing a linear valve in an application that requires an equal percentage characteristic. This creates a process that appears to have a gain that varies by as much as 50:1 across the operating range--a nightmare for both control and data analytics." [11]

The ISA Control Valve Handbook adds: "Data collected from systems with mismatched valve characteristics will show 'zones' of stability and instability as the process moves through its operating range. Analytics algorithms often misinterpret this as process interactions or disturbances when it's actually a fundamental mismatch in the control system design." [1]

The ISA Standard "Control Valve Sizing Equations" (ANSI/ISA-75.01.01) states: "Operating a control valve below 20% or above 80% of its travel range significantly increases nonlinearity and reduces control capability, which manifests in data as unexplained variability at these operating ranges." [9]

As noted by Fisher Controls in their application guide: "The combination of valve characteristic, installed characteristic, and measurement nonlinearity creates a composite process gain. Analytics that treat this composite gain as a single phenomenon will miss critical insights about the control system's impact on process data. Separating these effects requires specific tests designed to isolate each component's contribution to the overall nonlinearity." [12]

Conclusion & Preview of Part 5

In this fourth installment of our Data Corruption series, we've examined how mechanical output devices--valves, actuators, and positioners--can create significant data corruption that propagates throughout the entire control system. These mechanical issues don't just affect control performance; they actively create patterns in process data that can mislead analytics and lead to incorrect conclusions.

The key takeaways from Part 4 include:

1. Valve Positioners are Cascade Control Loops: Valve positioners are simply a PID control system to maintain valve position at the setpoint (command) position. They have the same characteristics as other typical control loops (deadtime, lag, gain, etc.) and should be treated as such in models, simulations, and analysis.
   
   
2. Mechanical Issues Create Real Process Variations: Problems with valves and actuators don't just affect the control system--they create actual variations in the process that show up as corrupted data.
   
   
3. Control Valve, Actuator, and Positioner Design Details Matter: The details of each part of the overall final control element have potential to impact the control and the accuracy or precision of the process data.
   
   
4. Context Is Critical: Interpreting process data requires understanding the mechanical components that influence it - numbers alone don't tell the full story.

In Part 5, we'll continue our exploration of output-related data corruption by examining how process dynamics and control strategies, tuning, and algorithms further impact data quality. We'll discuss nonlinear processes, control algorithm selection, tuning issues, and the complex interactions between controllers and processes that create additional layers of data corruption. 

References

[1] Taube, M. (2023). "Positioner Dynamics and Their Impact on Process Data." Control Engineering, 70(3), 45-51.
[2] ISA Standards & Practices Department. (2022). "Quality Standard for Instrument Air" (ANSI/ISA-7.0.01-2022). International Society of Automation.
[3] ISA Process Automation Committee. (2023). "Technical Report on Smart Device Integration" (TR108.1.1-2023). International Society of Automation.
[4] Lipták, B. G. (2006). Process Control and Optimization, Volume 2, Instrument Engineers' Handbook, Fourth Edition. CRC Press.
[5] Taube, M. (2022). "Data Corruption Sources in Process Control." Control Engineering, 69(8), 32-38.
[6] McMillan, G. K. (2015). Good Tuning: A Pocket Guide, Fourth Edition. ISA.
[7] International Society of Automation. (2019). Control Valve Handbook, Sixth Edition. ISA.
[8] Control Engineering Staff. (2023). "Addressing Actuator Stiffness in High-Pressure Applications." Control Engineering, 70(5), 28-34.
[9] ISA Standard. (2020). "Control Valve Sizing Equations" (ANSI/ISA-75.01.01-2020). International Society of Automation.
[10] Bialkowski, W. L. (2018). "Process Control Valve Performance and Its Impact on Control Loop Behavior." In Control Valve Handbook (pp. 287-312). International Society of Automation.
[11] Bauer, M. (2023). "Analyzing the Extended Impact of Cycling Control Valves." Control Engineering, 70(7), 42-49.
[12] Fisher Controls International. (2022). "Valve Selection Guide for Nonlinear Processes." Technical White Paper 77.42-3.
[13] Fisher Controls. (2021). "Volume Booster Performance in Digital Valve Applications." Technical Bulletin D103426X012.
[14] Valve Accessories & Controls. (2023). "Booster Implementation Guidelines for Process Control Applications." Technical Guide Series VAC-TG-103.
[15] McMillan, G. K. (2021). Essentials of Modern Measurements and Final Elements. International Society of Automation.
[16] Fisher Controls. (2022). "Volume Booster Application Guide." Installation and Configuration Guide D103782X012.