Sweet Smell of Success: HMcM and Yorkshire Water Services.

Detail: Where there's muck, there are consultants... Left to right, Simon Konn, Matthew Sheppee, Stewart Williams and Ian Richardson (ICS) savour the delights of a Yorkshire Water STW on a cold morning.

Sewage Treatment Works (STWs) are probably not generally seen as being at the glamorous end of the workspace spectrum. Consequently, the invitation to bid for a project that started with a tour of a selection of Yorkshire's finest sewage treatment works might not seem a particularly attractive one. However, contrary to this initial expectation, the project Hartley McMaster are just completing for Yorkshire Water Services (YWS) has been one of the most interesting and enjoyable pieces of work that the company has undertaken.

The Background: The Environment Agency sets capacity and performance levels the STWs have to satisfy, including flow rates through key parts of the system. The water companies are responsible for collecting this flow data and demonstrating that each site is operating within the agreed limits. To gather reliable data requires the onsite flow meters to be working accurately.

The Problem to be solved was how to spot if the flow meters required attention, given the variable patterns of flow, in part due to the vagaries of the weather. Traditionally, skilled engineers had studied the data and used their knowledge and experience to identify unusual behaviour which might indicate a system fault. However, as the number of monitored sites increased, the idea of developing analytical methods to assist the engineers became more attractive to all concerned. Hence the project which HMcM bid for and won.

The Solution: Developing parametric models of behaviour for the flow meters was considered a non-starter for this project. Instead the team set about developing robust techniques for spotting atypical behaviour in noisy data. Taking a large database of historical data - which included many examples of known flow meter failures - three techniques were applied, Statistical Analysis, Pattern Recognition by Neural Networks, and a Rule-Based Approach.

The Results: The results from each approach were reviewed and used to identify and refine the more promising approaches. These in turn were further investigated leading to the final recommendation of a combined statistical and rules-based method. This solution has been shown to detect anomolous behaviour (suggesting meter failure) with a level of accuracy above that required by the Environment Agency. Yorkshire Water Authority will now be implementing the approach into their flow data logging system, automatically generating alerts for the engineers who are responsible for maintaining the flow meters at the 500 STWs run by the company.

For more information about this project and Hartley McMaster's analytical projects in general, please contact Dr Mathew Davies, Email: daviesm@hmcm.co.uk