The safety of power transmission systems in wind turbines is crucial to the wind turbine’s stable operation and has attracted a great deal of attention in condition monitoring of wind farms. Many different intelligent condition monitoring schemes have been developed to detect the occurrence of defects via supervisory control and data acquisition (SCADA) data, which is the most commonly applied condition monitoring system in wind turbines. Normally, artificial neural networks are applied to establish prediction models of the wind turbine condition monitoring. In this paper, an alternative and cost-effective methodology has been proposed, based on the group method of data handling (GMDH) neural network. GMDH is a kind of computer-based mathematical modelling and structural identification algorithm. GMDH neural networks can automatically organize neural network architecture by heuristic self-organization methods and determine structural parameters, such as the number of layers, the number of neurons in hidden layers, and useful input variables. Furthermore, GMDH neural network can avoid over-fitting problems, which is a ubiquitous problem in artificial neural networks. The effectiveness and performance of the proposed method are validated in the case studies.

Strain sensor placement strongly influences the accuracy and cost of deformation reconstruction with the inverse finite element method (iFEM). This study develops a common sensor layout for several loading regimes by coupling the four-node inverse shell element with a Lichtenberg-inspired local search embedded in the Non-dominated Sorting Genetic Algorithm III (LA-NSGA-III). At a prescribed sensor count, the layout is optimized by simultaneously minimizing reconstruction residuals for multiple training loads. The method is assessed using DTLZ benchmarks and a plate example under equal objective-evaluation budgets, with separate controls for the multi-load formulation and the local search operator. Across 20 paired plate runs with 23 sensor sites, the proposed search reduces the mean joint training residual by 18.65% relative to NSGA-III, with mean displacement errors of 0.73% and 0.77% for the two training loads. Additional pressure, torsional and off-center force cases, a withheld combined load, strain noise, and sensor loss are used to assess a fixed layout under changing conditions. A cantilever plate test demonstrates strain-based displacement reconstruction, with laser peak displacement measurements used to calibrate the FE reference model. The method supports the offline design of a single sensor layout for iFEM reconstruction under multiple anticipated loading conditions, reducing the need to reposition sensors when the load changes.

Offshore wind-turbine gearboxes operate under continuously varying speeds, while incipient planetary-gear fault signatures are weak, modulated, and difficult to recover without a tachometer. This study presents a tachometer-free framework that separates low- and high-oscillation components using adaptive oscillatory behavior signal decomposition, with its Q-factor and sparsity parameters selected by a convergence index and multi-scale periodic-modulation-intensity double loop. The integrated workflow uses existing gearbox vibration channels for phase estimation and order analysis when a tachometer signal is unavailable. Its contribution is the coupling of adaptive component selection, compact high-order time–frequency analysis and ridge-based resampling. A downsampled fourth-order synchrosqueezing transform (DSST4) produces a compact, concentrated time–frequency representation. Penalized global-path ridge extraction then estimates shaft speed, and segmented integral-approximation resampling converts the signal to the angle domain. For a simulated amplitude- and frequency-modulated signal at −5 dB, DSST4 reduces Rényi entropy from 7.35 to 5.65 and computation time from 4.959 s to 0.002 s, relative to the full-sampling fourth-order transform. In the University of Ottawa variable-speed bearing dataset, speed-estimation error is 18.67 rpm. For a 2 MW offshore wind turbine, order spectra identify a ring-gear component near order 0.714 in March, and an additional planet-gear component near 0.456 with modulation sidebands in April. Teardown confirmed spalling on both gears.

Accurate bearing prognostics require health representations that distinguish persistent degradation from transient vibration responses. This study develops a sequential framework comprising a mechanism-informed gated health indicator (MIG-HI), causal evidence-driven transition identification (CETI) and a stage-conditioned dual-branch long short-term memory network (SCDB-LSTM). MIG-HI groups multidomain vibration descriptors into structural, energy, impulsiveness, fault mechanism, and cross-channel evidence and regulates impulsive contributions using mechanism and channel support. CETI combines level, trend and persistence evidence to establish a formal first prediction time (FPT). This reference links an offline degradation-aligned normalized prognostic target with a causally updated stage coefficient. SCDB-LSTM combines branch-specific representations through a shared temporal encoder and conditions regression on the stage coefficient. The evaluation described here includes condition-specific prediction experiments under three PHM 2012 operating conditions and detailed health-indicator analyses on five Condition 1 bearings. MIG-HI achieves the highest trendability among the compared indicators (0.4967), although it does not dominate monotonicity or prognosability. The target is dimensionless and does not directly estimate remaining operating time. Cross-condition transfer, disturbance-specific robustness and calibrated predictive uncertainty require further validation.