Underwater source localization problems are complicated and challenging: a) the sound propagation speed is often unknown and the unpredictable ocean current might lead to the uncertainties of sensor parameters (i.e. position and velocity); b) the underwater acoustic signal travels much slower than the radio one in terrestrial environments, thus resulting into a significantly severe Doppler effect; c) energy-efficient techniques are urgently required and hence in favour of the design with a low computational complexity. Considering these issues, we propose a simple and efficient underwater source localization approach based on time difference of arrival (TDOA) and frequency difference of arrival (FDOA) measurements, which copes with unknown propagation speed and sensor parameter errors. The proposed method mitigates the impact of the Doppler effect for accurately inferring the source parameters (i.e. position and velocity). The Cramér-Rao lower bounds (CRLBs) for this kind of localization are derived and, moreover, the analytical study shows that our method can yield the performance that is very close to the CRLB, particularly under small noise. The numerical results not only confirm the above conclusions but also show that our method outperforms other competing approaches.

Knowledge of the bubble-forming regime is important for studying the mechanism of gas-liquid two-phase flow. The process of bubble formation through an underwater nozzle at the bottom of a rectangular tank was observed using a high-speed camera. Five bubble-forming regimes were classified by considering both the bubble interaction and the periodic behavior. To automatically identify the bubble-forming regimes, 22 image textural features were extracted based on the grey-level co-occurrence matrix and grey-level gradient co-occurrence matrix. A feature selection method, the mathematical correlation weight algorithm (MCWA), was proposed to select the optimal textural feature subset for bubble-forming regime recognition. Experimental results demonstrated that the identification rates based on the feature subset selected by MCWA were 99.81% and 99.24% for support vector machine and artificial neural network, respectively, which were higher than those by the between-class distance method and the Fisher algorithm. The MCWA proposed in this paper is an effective feature selection method and can be widely used in pattern recognition.

To acquire the distribution of gas–liquid two-phase flow accurately, 3-D reconstruction method of bubble flow field with multiview cameras was studied in this work. Three cameras were employed to take bubble images from different angles, and the three-camera stereo calibration method is proposed to obtain the mapping relationship between bubble flow field and projection image. Then, an improved technique named rough-precise location algebraic reconstruction technique (ART) based on multiview images was developed to reconstruct the bubble flow field. The calibration result is used for rough location of bubbles and the ART is employed for precise location. Based on the simulation experiment, the mean Intersection-over-Union of the proposed method is 93.84% and the mean reconstruction time is 0.081 s/voxel, which are faster and more accuracy than traditional ART. The reconstruction experiments of bubble flow field have illustrated that the proposed method is of less noise and more accurate than the parallax method. Meanwhile, the proposed method solves the bubble overlapping problem to some extent.

: In recent years, multivariate sensor time series forecasting has become increasingly critical in industrial applications such as the condition monitoring of reciprocating compressors. However, the inherent non-stationarity, heterogeneity, and multi-period characteristics of industrial data pose significant challenges to traditional forecasting models. The offline training mode is difficult to adapt to the changes of dynamic working conditions, and the storage and calculation costs are high. Although continual learning (CL) provides the possibility of dynamic update, the existing methods have limited their applicability in industrial scenarios due to the low efficiency of random sampling and the unresolved stability-plasticity dilemma. To address these issues, this paper proposes a multi-scale dual-memory CL framework. The proposed method captures complex spatio-temporal dependencies through multi-scale modeling for CL. It improves replay efficiency and mitigates catastrophic forgetting by intelligently selecting and preserving representative samples via a dual-memory replay mechanism. A feature distribution alignment strategy is employed to reduce distributional shifts across tasks. In addition, an adaptive loss weighting mechanism dynamically balances the stability of old knowledge and the plasticity required for new tasks. The superiority of the proposed method and the effectiveness of each module are verified by comparative experiments and ablation experiments.