This thesis addresses the problem of PID controller autotuning based on an open-loop process step response, with the aim of optimizing its speed. The work therefore analyzes how non-equilibrium conditions, response truncation, and the choice of identification method affect the quality of the achieved tuning. The motivation is that in industrial practice, autotuning has often to be performed in conditions where it would be difficult or inconvenient to wait for the process to reach a steady state before applying the stimulus (initial non-equilibrium) and/or to detect that the response has come to an end (truncation). The analysis focused on three interconnected aspects: the initial equilibrium quality, the identification method, and the tuning law. As commonly done in autotuning, the process was approximated by a First Order Plus Dead Time (FOPDT) model, estimated through the Method of Areas (MoA), the Method of Moments (MoM), and subsequently the Method of Tangents (MoT). In addition, fast identification approaches were developed based on transient truncation and inflection point analysis, where a Newton type iterative procedure allows real time estimation of the parameters ˆ T, ˆD , and ˆμ. These methods significantly reduce identification time but introduce a trade-off between speed and model fidelity, which was analyzed through simulations and quantitative comparisons. Numerical and experimental tests were conducted on monotonic, oscillatory, and righthalf- plane zero systems to evaluate the influence of non-equilibrium and identification window length using ITAE, PM, and Ts as performance indices. The Cohen-Coon and Internal Model Control (IMC) tuning laws were employed to quantify the propagation of identification errors to controller design. The results show that premature autotuning generates systematic distortions: ˆ T tends to be underestimated and ˆD overestimated, with errors exceeding 10% for |y˙0| > 0.02 in oscillatory cases. Response truncation and the use of the inflection-point method further accelerate identification but reduce model fidelity, altering the ˆD/ ˆ T ratio and closed-loop behavior. The MoM proved to be the most accurate method for monotonic systems, while the MoA exhibited the greatest generality; the IMC rule showed the highest tolerance to identification bias. Overall, nonetheless, the study highlights a practical trade-off between speed and resilience: autotuning can be accelerated within defined limits by selecting method–tuning combinations consistent with the process dynamics.
Il presente lavoro di tesi affronta il problema dell’autotuning dei controllori PID a partire dalla risposta al gradino del processo in anello aperto, con l’obiettivo di ottimizzarne la velocità. Il lavoro analizza quindi come le condizioni di non equilibrio, la troncatura della risposta e la scelta del metodo di identificazione influenzino la qualità del tuning ottenuto. Nella pratica industriale, l’autotuning viene spesso eseguito in condizioni in cui risulta difficile o poco conveniente attendere che il processo raggiunga lo stato stazionario prima di applicare lo stimolo (non equilibrio iniziale) e/o rilevare il completo termine della risposta (troncatura), al fine di ridurre i tempi di messa in servizio. L’analisi ha considerato tre aspetti interconnessi: la qualità dell’equilibrio iniziale, il metodo di identificazione e la legge di tuning. Il processo è stato approssimato con un modello FOPDT (First Order Plus Dead Time), stimato mediante il Method of Areas (MoA), il Method of Moments (MoM) e, successivamente, il Method of Tangents (MoT). Sono stati inoltre sviluppati approcci rapidi di identificazione basati sulla troncatura del transitorio e sull’analisi del punto di flesso, nei quali una procedura iterativa di tipo Newton consente di stimare in tempo reale i parametri ˆ T, ˆD e ˆμ. Tali metodi riducono sensibilmente il tempo di identificazione, ma introducono un compromesso tra velocità e fedeltà del modello (model fidelity), analizzato tramite simulazioni e confronti quantitativi. Le prove numeriche e sperimentali, condotte su sistemi monotoni, oscillatori e con zeri a parte reale positiva, hanno valutato l’influenza del non equilibrio e della durata della finestra di identificazione utilizzando gli indici ITAE, PM e Ts. Le leggi di sintesi Cohen-Coon e Internal Model Control (IMC) sono state impiegate per quantificare la propagazione degli errori di identificazione sul progetto del controllore. I risultati mostrano che l’esecuzione prematura dell’autotuning genera distorsioni sistematiche: ˆ T tende a essere sottostimato e ˆD sovrastimato, con errori che per |y˙0| > 0.02 superano il 10% nei casi oscillatori. La troncatura della risposta e l’uso del punto di flesso accelerano ulteriormente l’identificazione ma riducono la fedeltà del modello, alterando il rapporto ˆD/ ˆ T e il comportamento del sistema in anello chiuso. Il MoM si è dimostrato il più accurato nei sistemi monotoni, mentre il MoA ha mostrato la maggiore generalità; la legge IMC garantisce la più elevata tolleranza ai bias di identificazione. Complessivamente, lo studio evidenzia un compromesso pratico tra velocità e robustezza: l’autotuning può essere accelerato entro limiti definiti, adottando coppie metodo–tuning coerenti con la dinamica del processo.
Fast PID autotuning from truncated step responses accounting for non-equilibrium initial conditions
Verga, Federico
2024/2025
Abstract
This thesis addresses the problem of PID controller autotuning based on an open-loop process step response, with the aim of optimizing its speed. The work therefore analyzes how non-equilibrium conditions, response truncation, and the choice of identification method affect the quality of the achieved tuning. The motivation is that in industrial practice, autotuning has often to be performed in conditions where it would be difficult or inconvenient to wait for the process to reach a steady state before applying the stimulus (initial non-equilibrium) and/or to detect that the response has come to an end (truncation). The analysis focused on three interconnected aspects: the initial equilibrium quality, the identification method, and the tuning law. As commonly done in autotuning, the process was approximated by a First Order Plus Dead Time (FOPDT) model, estimated through the Method of Areas (MoA), the Method of Moments (MoM), and subsequently the Method of Tangents (MoT). In addition, fast identification approaches were developed based on transient truncation and inflection point analysis, where a Newton type iterative procedure allows real time estimation of the parameters ˆ T, ˆD , and ˆμ. These methods significantly reduce identification time but introduce a trade-off between speed and model fidelity, which was analyzed through simulations and quantitative comparisons. Numerical and experimental tests were conducted on monotonic, oscillatory, and righthalf- plane zero systems to evaluate the influence of non-equilibrium and identification window length using ITAE, PM, and Ts as performance indices. The Cohen-Coon and Internal Model Control (IMC) tuning laws were employed to quantify the propagation of identification errors to controller design. The results show that premature autotuning generates systematic distortions: ˆ T tends to be underestimated and ˆD overestimated, with errors exceeding 10% for |y˙0| > 0.02 in oscillatory cases. Response truncation and the use of the inflection-point method further accelerate identification but reduce model fidelity, altering the ˆD/ ˆ T ratio and closed-loop behavior. The MoM proved to be the most accurate method for monotonic systems, while the MoA exhibited the greatest generality; the IMC rule showed the highest tolerance to identification bias. Overall, nonetheless, the study highlights a practical trade-off between speed and resilience: autotuning can be accelerated within defined limits by selecting method–tuning combinations consistent with the process dynamics.| File | Dimensione | Formato | |
|---|---|---|---|
|
Executive_Summary_Verga.pdf
accessibile in internet per tutti
Dimensione
835.29 kB
Formato
Adobe PDF
|
835.29 kB | Adobe PDF | Visualizza/Apri |
|
Thesis_Verga.pdf
accessibile in internet per tutti
Dimensione
3.67 MB
Formato
Adobe PDF
|
3.67 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/246654