Flow measurement in partially filled pipes presents greater complexity compared to fully filled systems, primarily due to the complex velocity distribution within the cross-section, which is a key source of measurement inaccuracy. To address this challenge, an ultrasonic flow meter was designed and developed, capable of simultaneously measuring both flow velocity and fluid level. To improve measurement accuracy, a flow profile correction factor (FPCF) was derived based on the velocity distribution characteristics and applied to the raw flow meter output. A dedicated open-channel flow loop incorporating a 250 mm diameter pipe was constructed to test and calibrate the system under controlled conditions. Flow rates in the loop varied from 2 to 6 liters per second. The accuracy of the flow meter was evaluated using the Flow-Weighted Mean Error (FWME) metric. Experimental results showed that applying the FPCF significantly improved accuracy, reducing the maximum flow measurement error from 8.51% to 2.44%. Furthermore, calibration led to a substantial decrease in FWME from 1.78% to 0.08%, confirming the effectiveness of the proposed methodology. The flow meter was also subjected to clogging scenarios by artificially obstructing the flow. Under these conditions, the device was able to reliably measure the flow and successfully detected the clogging, triggering an alarm to the operator to take necessary action.
4 August 2025 in IEEE Transactions on Instrumentation and Measurement
Transit time ultrasonic flowmeters offer a variety of advantages over other flowmetry methods; including low pressure loss, no moving parts, self-diagnosing capabilities, and possibility of hot-tap installation. In a transit time ultrasonic flowmeter, the average flow velocity is calculated from the upstream and downstream time-of-flights (ToFs) of the sound. To improve the ToF measurement, a novel signal processing method named pattern matching is introduced in this article. In this method, the wave packet is compared with a trainable pattern to measure the ToF. An ultrasonic flowmeter setup is designed to record the wave signals on a computer as an offline dataset. The performance of the pattern matching method is evaluated by its comparison with common classic and state-of-the-art algorithms by implementing them on the same dataset. The standard deviation (SD) of the time-flight in zero flow is used as the performance criterion. The results demonstrate that the pattern matching method (average SD $= 0.87~\mu $ s) exhibits the best performance compared to various threshold methods and cross correlation method. Ultimately, the pattern matching method along with state-of-the-art algorithms are implemented on an embedded hardware in a flow test setup. Only pattern matching method was able to follow the flow rate profile in high noise environment and had significantly lower error and better repeatability compared to other methods (error reduced from 14.1% to 0.322% compared to the second best performing cross correlation method).
Time-of-flight (TOF) measurement is the core functionality of signal processing units of ultrasonic flowmeters and distance meters. The received ultrasonic wave packet signal has a bell-shaped envelope, with an onset often obscured by noise. Various digital signal processing methods have been proposed to improve TOF measurement accuracy, employing either a limited set of selected data points from the received or conditioned signal, or all the data points within a defined segment of the received signal. The cross-correlation, model fitting and the pattern matching methods are prominent among those that leverage full-data approaches. This study introduces and evaluates two batch training strategies in pattern matching method: parametric modeling and optimized averaging. An offline dataset of 8000 received signals was recorded to implement the batch training methods. The first 4000 signals were used as the training data, and the subsequent 4000 were used as the test data. The conventional packet-envelope threshold and cross-correlation methods were implemented as baselines, yielding average errors of 9.10% and 4.32% and standard deviations of 2.38 μs and 1.95 μs at zero-flow respectively. In comparison, the optimized average method and the parametric model method (using the extended model) were able to reduce the standard deviation in the zero flow to 0.52 μs and 0.19 μs, and the average flow measurement error to 5.53% and 3.28% respectively.