Researchers at Czech Technical University in Prague have turned a familiar control-room procedure into a deterministic tuning algorithm. SPIN, short for Step-response Phase-portrait INspection, adjusted deliberately mistuned proportional-integral-derivative controllers on four simulated process plants until every test met the authors’ damping limits. It needed one closed-loop step response at a time, but no mathematical model of the plant and no optimisation solver.
PID controllers hold variables such as temperature, pressure and flow near a target. The proportional term reacts to the current error, the integral term accumulates past error, and the derivative term responds to how quickly the error is changing. Engineers often tune these three gains by changing the target, watching the resulting rise and oscillation, and adjusting one gain. SPIN converts that inspection routine into repeatable measurements.
For each step response, the algorithm builds three phase portraits, which plot one changing signal against another. They pair the control error with its accumulated value, its first change and its second change. SPIN then counts how far each trajectory winds around its settling point. The three dimensionless turn counts correspond to low-, middle- and high-frequency oscillation associated with the integral, proportional and derivative bands. A fixed decision rule lowers the gain responsible for the first certain violation and raises gains whose bands remain within their limits.
The researchers tested that rule on four lag-and-delay models representing lag-dominant, balanced, delay-dominant and high-order processes. Every run used the same thresholds and gain step, while each noise-free response contained 500 samples over ten apparent time constants. The starting proportional and integral gains were halved from an AMIGO design, and the derivative gain was tripled. At the returned settings, every turn count was inside its assigned limit. The integral gain more than doubled on all four plants, the derivative gain fell on all four, and no gain reached the edge of the permitted range.
Those adjustments remain visible to an operator: each decision can be traced to one of the three portraits and its turn count. That transparency distinguishes SPIN from a black-box optimiser and could make it useful as a starting point when a plant model is unavailable or expensive to maintain. The team has released a browser-based reference implementation that reproduces the simulated examples and exposes each gain change.
The evidence is still limited to an unreviewed preprint and four simulated plants. SPIN assumes a stable, self-regulating process whose gain and phase decline with frequency; integrating processes fall outside the present method. The tests used noise-free records, while real sensors can corrupt the differences used in two portraits, and filtering that noise remains future work. Each iteration also requires a response to settle. In one deliberately difficult high-order case, the first acceptable setting took 32 iterations. Hardware trials must therefore show whether the method can tune a working plant without excessive test time or disturbance.
Related coverage
Sources
- Daniel Pachner and colleagues, “Model-Free PID Tuning by Step-Response Inspection,” arXiv, September 14, 2026. Abstract, authorship, submission date, method scope and public implementation.
- Full paper. Phase-portrait construction, decision rule, plant battery, returned gain multipliers, hard-start test and limitations.
- SPIN interactive reference implementation. Public demonstration of the tuning procedure and reported simulations.
- Lead image: PEO ACWA, via Wikimedia Commons, CC BY 2.0. Reused unmodified as an illustrative photograph; it does not show the reported study.
.jpg){kind=link}