IEEE Transactions on Instrumentation and Measurement · 2023 · 52 citations · 27 references
EngineeringIndustrial EngineeringMechanical EngineeringAcousticsSensor FusionSensorsAcoustic ModelingAcoustic SensorsFlow MeasurementPhysical AcousticsPressure MeasurementPressure SensorsConvolutional Neural NetworksElectrostatic SensingMass Flow MeasurementMass Flow RateMulti-sensor Data FusionMulti-sensor FusionMultisensor FusionSound Pressure Level Measurement
Online continuous measurement of mass flow rate of pneumatically conveyed solids in square pipes is needed, but single‑sensor methods fail due to complex dynamics caused by the pipe’s four corners. This study proposes a multi‑sensor fusion and data‑driven modeling approach to overcome these limitations. A system combining acoustic, capacitive, and electrostatic sensors captures sound pressure, volumetric concentration, and velocity, from which statistical features are extracted and fed into a CNN‑LSTM model that is benchmarked against ANN, SVM, CNN, and LSTM models on a laboratory rig with velocities 11–23 m/s and flow rates 8–26 kg/h. The CNN‑LSTM achieves a relative error within ±1 % across all test conditions, outperforming all competing models.
Online continuous measurement of the mass flow rate of pneumatically conveyed solids in a square-shaped pipe is desirable for monitoring and optimizing industrial processes. However, existing techniques using a single type of sensor have limitations in measuring the mass flow rate of solids because of the complexity of the dynamics of solids flow due to the four sharp corners of a square-shaped pipe. This paper proposes a multi-sensor fusion and data-driven modelling-based method to tackle this challenge. A multi-sensor system based on acoustic, capacitive, and electrostatic sensing principles is designed and implemented to obtain the sound pressure level in the flow, volumetric concentration of solids, and solids velocity, respectively. Simultaneously, a range of statistical features is obtained by performing time-domain, frequency-domain, and time-frequency domain analyses on all sensor signals. The statistical features reflecting the variation of the mass flow rate of solids, as well as solids velocity and volume concentration of solids, are then fed into a data-driven model. A data-driven model based on a combined convolutional neural network and long short-term memory (CNN-LSTM) network is established, and its performance is compared with those of the back-propagation artificial neural network, support vector machine, CNN, and LSTM models. Experimental tests were conducted on a laboratory-scale rig on both horizontal and vertical pipelines to train and evaluate the CNN-LSTM model with solids velocity ranging from 11 to 23 m/s and the mass flow rate of solids from 8 to 26 kg/h. The CNN-LSTM model outperforms all other models with a relative error within ±1% under all test conditions.
27
Forecasting monthly gas field production based on the CNN-LSTM model
Wenshu Zha, Yuping Liu, Yujin Wan et al. · Energy · 2022 · 400 citations
Computing, Data Science, Forecasting +13