Quality indicators are critical for process monitoring and safety assessment in industrial chemical processes, yet their online measurement remains difficult or cost-prohibitive. This work presents a transparent and accurate soft sensor development framework termed the Dynamic Symbolic Regression Network (DSRN), which explicitly reconstructs symbolic relationships between easy-to-measure process variables and target quality indicators. Unlike mainstream black-box deep learning approaches, the proposed method leverages symbolic regression techniques to derive interpretable mathematical expressions. A genetic algorithm is employed to search for multi-layer optimal symbolic structures, while time-lag terms are incorporated to capture dynamic features, thereby revealing interpretable data-driven relationships within the chemical process. Furthermore, a fully objective feature selection strategy is introduced to determine the optimal subset of auxiliary variables by evaluating the uncertainty of the quality indicator via information entropy, which reduces the dimensionality of the DSRN input space and focuses the symbolic search on informative process variables. The effectiveness of the proposed approach is first demonstrated on a numerical example, where it accurately recovers the true variable relationships and achieves superior prediction performance. A case study on the Dow data challenge problem further validates that DSRN attains competitive prediction accuracy compared with advanced deep learning models, while offering explicit mathematical equations that provide valuable interpretability for operators. These results highlight the great potential of DSRN for practical process monitoring and control applications.
This study presents a conceptual multi-scale characterization framework for analyzing the total losses in oil-immersed ONAN distribution transformers. Rather than proposing a real-time partial discharge detection system or a validated energy recovery technology, our work integrates spectral energy density analysis, chemical and microstructural characterization of silicon steel laminations from failed transformers, infrared thermography, and multivariate statistical modeling. A chemical analysis revealed differences in carbon content associated with thermal faults (average 0.097%), partial discharges (average 0.061%), and electric arc failures (average 0.048%). X-ray diffraction and the Debye–Scherrer equation were used to determine the crystallite size, which was subsequently correlated with the total power losses through multiple linear regression. The resulting preliminary model (PT=140.810−0.255x1−1.846x2) showed reasonable agreement with experimental data, although its predictive capability is limited by sample size. The findings provide diagnostic indicators that complement conventional partial discharge assessment methods and establish preliminary relationships between microstructural degradation features and energy dissipation in distribution transformers. This multi-modal approach contributes to a better understanding of loss mechanisms and supports future research on condition monitoring and the material characterization of power transformers.
Force chains, which transmit and distribute loads through networks of interparticle contacts, largely determine the mechanical response of granular materials. However, automatically identifying force chains and quantifying their evolution remain difficult, especially for three-dimensional granular systems under fluid action. In this work, a coupled computational fluid dynamics and discrete element method (CFD-DEM) is combined with community detection to extract and characterize the force chain network of a granular packing during particle migration. The inter-particle contact forces are encoded as a weighted network, and an improved Louvain algorithm based on modularity optimization is employed to partition the force network into sub-communities. The evolution of the community structure is quantified in terms of community size, average contact force, centrality and coordination number at different fluid pressures and dynamic viscosities. The results show that strong force chains assemble into compact and highly branched communities located mainly in the central region of the packing, and that increasing fluid pressure promotes the recombination of strong and weak chains into more branched structures, enhancing the stability of the granular system under high pressure. The community size distributions display a linear decay in the complementary cumulative distribution function without a characteristic scale, and the coordination number distributions of different communities respond to pressure in opposite ways. Higher fluid viscosity retards particle migration and favors more compact and stable community structures. The proposed framework provides a quantitative link between the mesoscale organization of force chains and the macroscopic stability of granular materials under fluid action. Beyond the specific packing studied here, the framework relies only on the interparticle contact forces and can therefore be transferred directly to any numerical or experimental dataset in which such forces are accessible; it thus offers a globally applicable tool for diagnosing fines migration, permeability impairment and sand production in deeply buried reservoirs, as well as internal erosion in dams, filters and embankments and the handling of particulate beds in the chemical and pharmaceutical industries.
Hydrates formed during a subsea flowline shutdown can persist as deposits that restrict the subsequent restart. Assessing restartability therefore requires both thermodynamic conditions and the pressure needed to recover flow. This study develops a one-dimensional screening model that couples transient heat transfer, suspended and deposited hydrate inventories, local area loss, and a source-pressure projection. A representative 30 km deepwater flowline in the South China Sea is examined over production, shutdown, staged restart, and recovery. In the case without monoethylene glycol (MEG), the maximum deposited fraction increases from 0.0173 at 24 h to 0.549 at 60 h, reducing the minimum open-area ratio to 0.451. The initial reference pressure drop is 1.51 MPa, below the available 4.54 MPa. The subsequent ramp first activates the pressure limit 2.06 h after restart, and the minimum available-flow fraction is 0.333 at 8.5 h. The 30.2 and 56.5 m3 d−1 MEG cases have maximum subcoolings of −0.00759 and −4.13 °C, respectively, and remain hydrate-free in the calculations. Parameter perturbations show that deposition and resuspension affect restriction severity, while the assumed bubble diameter can change the restart classification. Latin hypercube sampling and a proper orthogonal decomposition surrogate describe conditional threshold exceedance. The results support comparison of shutdown, restart, and inhibition strategies under the specified model assumptions.
Sand production and proppant flowback in tight-gas reservoirs introduce gas–liquid–sand three-phase flow into wellbore and surface gathering systems, where particle-laden streams simultaneously cause mechanical erosion, electrochemical corrosion, and their synergistic wall-thickness loss at bends, tees, chokes, and separators. This study develops a novel field-informed prediction framework that, for the first time, couples gas–liquid–sand three-phase flow, orientation-dependent sand transport probability (STP), erosion–corrosion models with multi-size particle superposition, and Bow-Tie/Bayesian-network-based dynamic risk assessment into a single operating-envelope calculation. The key innovations include: (i) a unified erosion–corrosion coupling model that integrates mechanical erosion, electrochemical CO2 corrosion, and their synergistic interaction through particle-size-distribution-weighted superposition; (ii) a four-pipeline-type differentiation (inlet, gas outlet, sand outlet, water outlet) revealing that erosion–corrosion severity varies by more than a factor of three among pipeline types; and (iii) a Poisson–Gamma dynamic risk updating model that tracks failure probability evolution under time-varying operating conditions. The framework was applied to three tight-gas wells and their gathering pipeline network. Field sieve analysis constrained particle-size distributions (70–160 mesh), and the sand samples were confirmed as quartz-rich proppant (Mohs hardness 7) based on sieve analysis and visual inspection. Horizontal-section STP ranged from 76.25% to 91.43%, and vertical-section STP reached 100% in all wells. The predicted erosion–corrosion rate for the KQ51H pipeline was 1.866 mm/a, exceeding the critical threshold and classifying it as High risk. Model predictions of wall-thickness loss at 11 monitoring locations agreed with ultrasonic thickness measurements within 10% error. The Bow-Tie model identified sand mass rate, particle-size distribution, and corrosion-product film failure as critical risk drivers. Dynamic Bayesian updating tracked failure probability evolution over 12 months, with the highest-risk segment showing a probability increase from 0.0149 to 0.0325. The framework supports differentiated inspection scheduling and mitigation prioritization across pipeline segments in a gathering station.
