Pin arrangement significantly affects rubber-melt mixing and extrusion in pin-barrel cold-feed extruders. However, internal flow details are difficult to observe experimentally, and efficient prediction of flow-field parameters remains unavailable. This study used a finite-element model preliminarily validated against measured temperatures, together with particle tracing, to compare configurations with 0, 2, 4, and 6 pins per group. A dataset of 140 pin arrangements was generated by Latin hypercube sampling and numerical simulation. A mind evolutionary algorithm-optimized back-propagation neural network (MEA-BP) was then developed to predict melt volume-averaged temperature and average shear rate. Pins increased melt velocity and shear heating and improved cross-sectional temperature uniformity. Among the four uniform configurations, the 4-pin-per-group configuration showed the fastest reduction in segregation scale with a moderate residence time, achieving a favorable balance between mixing adequacy and processing efficiency. Particle tracing indicated repeated fluid splitting and recombination, whereas further increases in the number of pins yielded limited benefits. Under identical data partitions, network settings, and evaluation conditions, MEA-BP achieved R² values of 0.957 and 0.872 for temperature and shear-rate prediction, respectively, outperforming GA-BP, PSO-BP, and conventional BP.

Cross-condition fault diagnosis is important for ensuring the reliable operation of mechanical systems. However, most existing methods assume that the source and target domains share identical fault label spaces and data distributions, limiting adaptation to unknown faults and cross-condition shifts in practical industrial scenarios. To address this issue, a structured, reliable, domain-specific open-set domain adaptation method is proposed. The proposed method first constructs a domain-specific batch normalization-based feature extraction network, in which independent normalization branches model statistical discrepancies under different operating conditions; it then designs a cross-domain structured representation consolidation module to enhance feature discriminability through source-domain anchor compactness, target-domain multi-view contrastive, and prototype entropy regularization constraints; an open-set boundary learning mechanism is further introduced to establish a discriminative boundary between known and unknown classes; finally, a reliability-aware pseudo-label propagation strategy refines target-domain pseudo-labels and imposes separate prediction-consistency constraints on known and unknown classes. Experimental results on the CWRU bearing dataset and the self-built rolling bearing dataset show that the proposed method achieves average H-scores of 93.32% and 97.65%, respectively, on open-set transfer tasks. Compared with several baseline methods, the proposed method achieves a better balance between known-class recognition and unknown-class detection, thereby improving cross-condition open-set fault diagnosis performance.