Model & equations
A numerical tuning simulator for a stable first-order process with downstream transport delay. It does not connect to or write gains to equipment.
Mark VIe thermal heater
Reported target: 350°F; observed output-to-temperature dead time approximately 240 seconds. The valve controls heater energy. The process gain and time constant are not characterized. Four minutes is dead time, not the time constant.
The supplied help page specifies the parallel equation OUT = PG·e + (1/IG)∫e dt + DG·de/dt, where IG is reset time. The user confirmed the original PG was 1.6, IG was entered directly as 0.333, and IG uses seconds. Thus the original integral coefficient in block coordinates is 1/0.333 = 3.003003… s⁻¹, not 0.333 and not PG/IG. The preset uses these historical original coefficients, not a proposed retune. The later screenshot shows trial PG 0.825, IG 0.5 and DG 0; it is not the original baseline. Original derivative settings remain unspecified.
PV/error and output scaling are still unverified, so block coefficients are not yet engineering-unit simulator gains. The thermal preset leaves the process gain, time constant, baseline temperature and operating output blank. Simulation and PI suggestions stay blocked until the engineering-unit model and gains are supplied. Selecting engineering-unit gains clears the native coefficients; it does not perform a scaling conversion. Native drafts saved before this clarification have their ambiguous Ki entry cleared, while other saved entries remain. Selecting the thermal scenario reloads the original coefficients; ?preset=thermal-original explicitly opens that fresh baseline.
The photographed help also mentions an optional squared-error mode; the full configuration, PV/SP scaling, output scaling, execution period, derivative behavior and tracking options still need checking. The simulator uses linear error and filtered derivative of feedback; the displayed block equation alone does not establish every configured behavior.
Existing trend data
Export timestamp, temperature PV, temperature setpoint, heater valve command and valve feedback if available. Include ammonia flow, inlet temperature, mode and saturation flags when logged. Use a continuous period before and after normal output changes, including any settled intervals available. The app does not automatically identify a plant from an unsettled endpoint or prescribe an operating test.
With defensible steady endpoints, first-order gain is Δtemperature/Δheater output; the time constant is the time after the dead time to reach 63.2% of the final temperature change. Without endpoints, identification needs a fitted dynamic model and uncertainty assessment; no fitted model is currently claimed.
Smith compensation is feedback dead-time compensation. The user confirmed a straight-line conversion from flow 0–340 lb/hr to an FF signal of 0–450°F: FF = flow × 450/340. The curve calculator evaluates this supplied mapping only within the stated range; out-of-range behavior of the real block is unknown. It does not prescribe a flow, change the 350°F setpoint, or add a temperature-valued signal directly to actuator output. The feed-forward injection path and its relationship to heater energy still need to be modeled; no combined feed-forward heater simulation is claimed.
Process and predictor
τ dx/dt = K(u − u₀ + d) − x
y(t) = y₀ + x(t − L(t))
τ̂ dx̂/dt = K̂(u − u₀) − x̂
yp(t) = y(t) + x̂(t) − x̂(t − L̂(t))
The plant and model receive the actual saturated controller output. The input load d is unmeasured by the model. Deviation states and prehistory start at zero around the entered equilibrium operating point (y₀, u₀). The baseline is assumed steady; it is not an arbitrary initial condition for a non-equilibrium heater. The setpoint steps from y₀ at 1 second. Transport delay is applied to output history with linear interpolation; variable delay here is a time-varying observation/transport model, not a physical distributed pipeline or actuator-delay model.
Variable delay
Choose a constant, step, or nonnegative-clamped sinusoidal actual delay. “Track supplied delay” follows that known delay schedule, using the selected tracking lag; zero lag follows it exactly. “Fixed model delay” holds the nominal delay constant. This is not automatic identification of an unknown delay.
PID convention
e = r − yf
u = clamp(u₀ + Kp e + I − Kd Df, umin, umax)
dI/dt = Ki e
With Smith enabled, yf = yp. Otherwise yf = y. Df is a first-order filtered derivative of feedback, avoiding derivative kick on the setpoint. Kp, Ki and Kd are parallel gains with units u/y, u/(y·s), and u·s/y respectively, where u is actuator units and y is process units; for ideal PID with nonzero gains, Ti = Kp/Ki and Td = Kd/Kp. The baseline is a separate conventional PID loop with identical gains, plant, limits and disturbance, not a separately optimized PID.
Conditional anti-windup freezes integral updates only when they would push farther into saturation. Negative process gains are supported with appropriately signed controller gains. Output limits are actuator units; gain maps actuator units to process units.
PI starting point
“Suggest PI gains” uses a SIMC-style nominal-model starting point. With Smith enabled: Kp = τ̂/(K̂λ), Ti = min(τ̂, 4λ), Ki = Kp/Ti, Kd = 0. With Smith off, use λ + L̂ in place of λ. Larger λ gives slower, generally less aggressive control. This is a starting point, not an identified plant, guaranteed stable tuning, or a replacement for commissioning.
Metrics and numerical method
Metrics cover the complete simulated run from the setpoint step, including any later delay change or load disturbance. IAE is the time integral of absolute tracking error, in process-unit seconds. Peak overshoot is the greatest excess beyond the requested setpoint relative to its step magnitude from y₀ (undefined for an unchanged setpoint). Final error is setpoint minus final process value. IAE improvement compares the Smith loop against conventional PID using the same gains; negative values mean worse tracking.
The simulation normally uses a 20 ms controller sample; long runs over 600 seconds use 100 ms when both time constants are at least 5 seconds and the derivative filter is at least 0.5 seconds. This simulation sample is not the Mark VIe block execution period. It uses exact zero-order-hold first-order updates, and a discretized filtered derivative. Timing is simulated seconds, not browser wall-clock time. Very aggressive gains or a rapidly changing delay need finer-step validation outside this lab.
Storage and export
Settings save in this browser’s local storage. Save/load setup uses versioned JSON. CSV contains the complete run, including both controllers, predicted feedback, delays, and load. Nothing is sent to a control system.