Identification of mechanical rotor axis trajectory state based on photogrammetry technology

Introduction State monitoring of rotating machinery is crucial for ensuring industrial production safety, and the rotor axis trajectory is a key indicator reflecting its operating status. Traditional contact measurement methods are susceptible to factors such as electromagnetic interference, which poses a risk of data distortion. Methods A research proposes a mechanical rotor axis trajectory state recognition method based on photogrammetry technology. Firstly, to address the issue of redundant and unstable feature points in complex operating conditions of accelerated robust feature algorithms, local two-dimensional entropy is introduced to purify candidate points. Secondly, in response to the efficiency and global shortcomings of traditional machine learning models that rely on manual parameter tuning, a support vector machine state recognition model was constructed that integrates directional gradient histogram features and transient search optimization algorithm. The transient search optimization algorithm was used to globally adaptively optimize the classifier’s penalty factor and kernel parameters. Results The improved accelerated robust feature algorithm reduces the number of feature points by about 38% and reduces trajectory reconstruction error by nearly half. The transient search optimization support vector machine model constructed has an average classification accuracy improvement of about 3% and a prediction speed improvement of nearly three times compared to other comparative models. Discussion The proposed solution successfully constructs a complete link from high-precision visual trajectory acquisition to state recognition, which not only provides a new non-contact solution for traditional contact measurement problems, but also has important engineering application value for improving the predictive maintenance level of equipment.