Flow meter drift is a serious problem, the financial losses from which can be huge. Flow meters do not usually get much attention and a drift can go unnoticed for a long time. In this research, a novel method is presented for the early detection of flow meter drift. The method is based on statistical process control (SPC). The method can be used with any type of flow meter (ultrasonic, magnetic etc.) regardless of its manufacturer. Another advantage of the presented method is that it does not require any knowledge of the mathematical models and relations that govern the process. The method is also capable of working with minimal data: only the monthly billing data are needed. Adapting the process can be done inexpensively.

In this paper, artifi cial neural networks (ANNs) were used to assess the performance of fl ow meters used in industrial water supply. These fl ow meters are susceptible to drift, a condition causing them to give erroneous readings that are inconsistent with the actual fl ow. A simulation of industrial water fl ow to the industrial consumers was made. This simulation contained both healthy and drifting fl ow meter readings. ANN was built and trained on the simulated data. At the time of testing, the ANN developed was correct 89.52% of the time in determining the status of the fl ow recorded by a fl ow meter. Keywords: artifi cial neural network, fl ow meter drift, industrial water supply, statistical process control. INTRODUCTIONThe importance of water cannot be overemphasized. Because of this, people realized the importance of controlling, organizing, and regulating its use. No sensible controlling, organizing, regulating, and planning can be done without accurate measurement. Hence, the abundance of fl ow meters in any water system. Accurate measurement arrangements are required for effectively managing water infrastructure regimes. Flow meters can be subject to specifi c failures such as erroneous readings and breakdown. Inaccurate measurements affect the controlling, organizing, and regulating processes of the water system and planning. Measures for avoiding and mitigating such fl ow meter problems include: calibration, repair, and replacement. Flow meter calibration generally involves checking against accurate standards to determine any appreciable deviations and correcting for errors. Yet, this calibration is not a one time transaction as it does not end the problem forever.Flow meter calibration is often neglected by the operation and maintenance staff of a water system. Drift is an error in measurement, which can increase with time. Ben Salamah et al.[1] outlined a statistical process control (SPC)-based fl ow meter drift detection method for the industrial water use regime. Realizing the usefulness of artifi cial neural network (ANN) modeling for forecasting and classifi cation objectives, an extension of this research considered specifi c applications of ANN modeling with improved SPC frameworks for fl ow meters. This paper presents some key fi ndings from the application of ANN models for detecting fl ow meter drifts.

Large flowmeters are used in many industrial facilities, including power plants, cooling-water stations for refineries, and petrochemical plants. These flowmeters are employed for various purposes, including billing. Just like all machines, flowmeters are subject to failure. Drift is a particular type of failure in which the flowmeter produces an error in measurement that would incrementally increase with time. Maintenance technicians calibrate and fix all measuring equipment, including flowmeters. Nevertheless, downsizing policies and budget cuts in most contemporary industrial facilities have made these technicians overwhelmed with work. A mathematical and computer-based drift-detection scheme is developed to reduce the burden of the maintenance staff. The detection scheme only uses the flowmeter's flow data and the discrete Fourier transform (DFT). The detection scheme was applied over the flow data from an actual flowmeter that drifted during its operation. DFT application over the data produced by the flowmeter led to expected results and other unexpected results. This paper discusses both results and suggests areas for further study. Practically speaking, the scheme would facilitate the early detection of drifts in flowmeters having seasonal flow regardless of type or manufacturer.