Oscillation Detection in Control Loops using PRONY Analysis

Oscillations in process control loops present a serious threat to normal functioning of plant as they lead to product quality variations, fast wear and tear of plant components, possible loss of human lives and property. As a result, overall profitability of the plant gets compromised. To cater to these issues automatic detection of plant wide control loop oscillations is a top most priority for control engineers and industrial consultants. In this work a new automatic oscillation detection algorithm has been proposed based on Prony method for time domain design of infinite impulse response (IIR) filter. In this method IIR filter is designed with time domain impulse response same as the input data. Further, oscillations in the control loop were detected based on the location of IIR filter poles on z-plane. Performance of the proposed algorithm was evaluated for oscillation detection for simulated data, real plant data and industrial plant data. Proposed algorithm was successfully able to detect oscillations in all the cases.

I. Introduction

One of the widely occurring problems in process control loops are oscillation of process-variable (PV) around set-point (SP). Oscillations in control loops can arise due to several factors like improper tuning of controllers, disturbances, poor disturbance rejection capability of controllers, presence of inherent nonlinearities like stiction, dead-band, hysteresis in final control element (FCE) and process nonlinearities [1] – [7] . Oscillations have severe effect on the plant performance. They lead to the variation in product quality and increase the rejection ratio which ultimately reduces the profitability of the plant. Oscillations also accelerate wear and tear of plant components which restricts the plant from operating at optimal point. In case of safety, critical control loops oscillations can also lead to disastrous consequences and may result into loss of human lives and property [1] – [7] . Hence it is desirable to detect oscillations in control loops as soon as they appear. It is nearly impossible to detect oscillations in control loops just by manual observations as there could be thousands of control loops in any typical process industry and therefore it demands automatic oscillation detection algorithms [2 , 4 , 5 , 8] .