Relative pose measurement of uncooperative spinning targets enables on-orbit servicing, debris removal, and space situational awareness. Monocular methods recover the full three-degree-of-freedom (3-DOF) rotation from images even though the chaser's own navigation state already fixes much of it, so a low-resolution image re-measures known quantities at a ${\approx}3$--$5$~deg error floor. We adopt the principle ``measure only what the chaser does not already know'': for a gimbaled payload tracking the target, the relative attitude separates into an orbital-rotation component fixed by geometric priors (orbit, attitude, gimbal angles) and the target spin, the only DOF requiring the image. On simulated data: (i)~the full-geometry prior reproduces the ground-truth orbital rotation to below $2\times10^{-4}$~deg under error-free telemetry, its residual reducing to the target spin rate ($1.1985$~deg/frame); a tested $0.22$~deg/frame noise budget leaves the spin-rate estimate statistically unchanged; (ii)~the spin rate attains a Kalman-filtered mean absolute error of $0.18$--$0.43$~deg/frame ($0.46$--$0.51$ extended); the spin-axis tilt, from equal-weight fusion of three channels, falls from a $4.6$-deg eight-configuration mean on the $47$~deg arcs to $0.6$--$0.7$~deg on the $92$~deg extension of Seq.~2, the $47$~deg-arc tilt propagating to a $9.4\%$ full-rate error; (iii)~the position follows from a Clohessy--Wiltshire fit of the line-of-sight history under the near-circular prior, with no range input, to $0.13$--$0.32\%$. It is model-free and works at 256 px. The prior does not lower the per-frame composed rotation error (the homography floor) but provides attribution, the unobservable spin axis, and a $1.3$--$3.4\times$ drift reduction.
Reliable inspection of thermal protection tiles requires measurement techniques capable of maintaining diagnostic accuracy under variations in sensor coupling and damage configurations. However, ultrasonic guided-wave measurements obtained using mobile contact scanning often suffer from arrival-time fluctuation, waveform distortion, and distribution shifts caused by changes in defect geometry. This work presents a physics-guided measurement framework for robust debonding detection in multilayer thermal protection structures. A deployable pitch-catch scanning strategy using a removable transducer pair is developed, followed by a physics-guided signal standardization procedure to reduce coupling-induced temporal variability and isolate damage-sensitive wave packets. A domain-adaptive learning model is subsequently introduced to compensate for residual distribution differences associated with damage-geometry variations. Experimental measurements from thermal protection tile specimens demonstrate that the proposed method achieves 95.68% accuracy and a Macro-F1 score of 0.952 on physically held-out target specimens, outperforming the source-only CNN and the compared domain-adaptation methods. The proposed methodology provides a robust ultrasonic measurement solution for reliable inspection of reusable aerospace thermal protection structures under practical scanning conditions.
Measurement validation is often reported as a binary property of an instrument, although a single system may support some bounded conclusions while leaving stronger ones unsupported. This paper formulates validation as a constrained scientific claim-admissibility problem. A general operator maps declared intended use, evaluated evidence, claim-specific requirements and claim strength to admitted and maximal claims, blocked overclaims, retained limitations, abstentions and next evidence actions. The Evidence-Gated Decision Framework implements this operator through applicability resolution, strict non-compensation and a versioned audit trace. Formal consistency properties establish monotonicity, irredundant evidence bases and local degradation under evidence removal for a frozen model. Measurement uncertainty, traceability and JCGM 106-type conformity decisions remain locally evaluated evidence objects. A synthetic temperature demonstrator isolates the decision logic, while an independent real-data motor-vibration case adds practical evaluation using 24 OROS OR35 acquisitions at four rotational-speed conditions. A magnetically mounted Dytran 3214A2 accelerometer and a Monarch ROS optical tachometer are acquired on the same time base. Across 600, 800, 1000 and 1200 min−1, the absolute group-mean vibration-derived 1×/tachometer frequency differences range from 1.94 × 10−4 to 5.92 × 10−4 Hz, the maximum absolute record-level difference is 0.002408 Hz, and the expanded comparison uncertainties are about 0.0096 Hz (k = 2). The case shows how one local compatibility result supports a bounded frequency-consistency claim without implying traceable acceleration, universal spectral interpretation or field readiness. Seven claim-separable evidence classes are instantiated as EVL 0–6, with EVL 7 reserved for the adjudicated output.
Global Navigation Satellite System (GNSS) real-time kinematic (RTK) positioning can provide accurate and reliable location services. Previous studies have investigated RTK algorithms constrained by carrier velocity information, whereas limited attention has been paid to the resilient refinement of carrier velocity information and motion states. In this study, a resilient velocity-constrained RTK framework for GNSS RTK (RVC-RTK) is proposed. Specifically, the proposed RVC-RTK resiliently integrates the velocity estimates from both time-differenced carrier phase (TDCP) and time-differenced pseudorange (TDPR) methods. Then, a sliding window strategy is adopted to refine reliable multi-epoch velocity estimates and further classify the receiver motion states. On this basis, resilient velocity constraints are imposed according to different motion states. Experiment results reveal that the accuracy of TDPR velocity estimates is 0.234, 0.279, and 0.317 m/s in X, Y, and Z components, respectively, while a higher accuracy of 0.008, 0.011, and 0.008 m/s is achieved for TDCP velocity estimates. This demonstrates the feasibility of introducing velocity constraints to improve the RTK positioning. Compared to the traditional RTK algorithm, the proposed RVC-RTK framework improves the positioning accuracy by 38.1%, 53.8%, and 23.5% in east, north, and up components, respectively. In conclusion, the proposed RVC-RTK framework can effectively improve the RTK positioning performance in real kinematic applications.
In pumping unit well fault diagnosis, non-analytical indicator diagrams frequently occur due to the privacy of field data and historical accumulation, resulting in invalid information and pixel redundancy that reduce diagnostic efficiency. To address this, a serialization method is proposed to convert two-dimensional redundant diagrams into one-dimensional sequences, enhancing information density and constructing a serialized fault dataset. To further tackle low diagnostic accuracy and catastrophic forgetting in Few-Shot Class-Incremental Learning, a Few-Shot Class-Incremental fault diagnosis model for non-analytical indicator diagrams based on serialization and contrastive learning (CLIFS-Net) is proposed, which is specifically designed for the proposed special serialized data format. First, the self-supervised contrastive learning module is pretrained on base classes and then frozen, learning stable and transferable representations from serialized data by capturing local contour variations and global morphology. Then, a lightweight adapter and sample replay strategy, combined with a Nearest-Mean-of-Exemplars classifier, enable flexible adaptation to new classes and alleviate forgetting. Experimental results on the serialized non-analytical indicator diagrams dataset show that the proposed backbone and classifier respectively enhance baseline methods by 2.39% and 4.23%. The integrated CLIFS-Net achieves an average diagnostic accuracy of 92.72%, higher than the overall average of 87.94% for baseline, backbone-modified, and classifier-modified methods.
