This paper considers how new techniques in advanced multivariable monitoring can be utilised
to provide predictive diagnostics for waste water processes. The methodology described is a
hierarchical approach where initially the sensors own fault codes along with simple univariate
statistical tests, are used to isolate the more obvious sensor faults. The next level is to consider
multivariable fault detection techniques, where the correlations and inter-relationships
between process signals can be utilised to detect genuine process faults in the most timely
manner possible. As well as providing diagnostic information, multivariable models can be used
to produce “soft sensor” reconstructions of faulty or missing sensor data.
The final layer in the hierarchy is the utilisation of the validated process and diagnostic data, for
control, operator intervention, and maintenance planning. Case studies are used to illustrate
predictive diagnostics in action.
Predictive Diagnostics, Asset Monitoring, Multivariable Control, Sensor Quality
The wastewater industry faces considerable challenges in making sense of the real-time data
that is collected from its processes, and in using this data to the best possible effect. This
industry (like many others) has a pressing need to minimise its energy and operational costs,
whilst maintaining product (in this instance final effluent) quality. Coupled with this business
driver is an often unforgiving environment for process sensors. Very often, wastewater
operations staff have little tolerance for instrumentation that they see as ‘flaky’. While the last
10 years have seen dramatic improvements in the quality and robustness of online wastewater
instrumentation, the questions still to be answered now are:
This paper details our evolving methodology for multivariable diagnostics, sensor
reconstruction, and fault tolerant advanced control. Our approach is illustrated with case
studies from operational control and monitoring systems from the UK water industry.