Proportional, Integral and Derivative feedback control (PID) is a mature technology responsible for the majority of automated decision making in the process industry. Despite the high reliance on this technology, low levels of maintenance and performance measurement are the norm in the process industry. Several analysis techniques exist for identifying oscillation, and then highlighting the root cause of the problem. Several time and frequency domain statistical techniques, as well as wavelet analysis are used to diagnose loop performance. In this study, 127 different control loops are analysed, and in depth troubleshooting is performed on a selection of 18 different control loops. The performance of flow loop F1035 is tracked through a number of different analysis techniques, highlighting the pitfalls of using only a single analysis technique. Lower order statistics and minimum variance performance analysis show that the loop is performing well. Plotting the PV-OP relationship suggests non-linear tendencies on F1035, and this is corroborated using high order statistical analysis (bicoherence). Non-linear loop behaviour is often as a result of a slip stick cycle, a sign that valve maintenance may be required. Frequency (power spectrum) analysis shows a 43 minute dominant oscillation, suggesting a low frequency disturbance affecting loop performance. Process units are typically exposed to cyclic behaviour occurring at several different frequencies, each having a different effect on the control of the process. By using a frequency based approach based on sinusoidal basis functions (ie Fourier analysis), these different frequencies get aggregated. This smudging of specific frequency information makes it difficult to pin-point the root cause, and makes the grouping of common oscillations difficult. In order to address the above issue, F1035 is analysed using othornormal wavelet basis functions. The results show that the period of oscillation is affected between day and night, with roughly a 2 minute oscillation prevalent at mid night, compared to a 100 minute oscillation at mid day. Obviously the 12 hour day-night swing is also prevalent. This information is unique to this approach. Ways of visualising changes in oscillatory behaviour using the wavelet analysis are also presented. Technical analysis of controller performance is only a small subsection of the issues that need to be considered when implementing a loop monitoring and maintenance solution. Issues such as connectivity, configuration, analysis, reporting and auditing are key in designing a workable maintenance environment for PID loop maintenance. Several packages are available commercially to assist industry in performing loop maintenance. When evaluating which package is best suited to a specific requirement, it is important to consider several different issues. The different audiences with a vested interest in loop performance require special attention in terms of reporting requirements. Visualisation of results is often more important than the physical measure of performance. Finally, the ability of a company to benchmark itself against current best practices and performance is often perceived as a major advantage. The results presented and discussed were generated using real industrial data. Information regarding suggested best practice when evaluating commercially available products is based largely on the author’s personal experience in the large scale industrial installation of such a monitoring solution.
Dissertation (MEng (Process Control))--University of Pretoria, 2007.