With the development of marine science construction, the application of submersibles is developing towards deep-sea, multi-functional, long-range, and ultra-high speed directions. Integrated navigation systems are widely used in deep-sea navigation due to their high reliability and good real-time performance. With the increasing scale of submersibles and the complexity of mission environments, the failure rate of navigation systems is also increasing. Once a failure occurs, it will lead to system failure or even mission interruption. Therefore, it is necessary to establish a comprehensive fault detection scheme, timely detect faults and make corrections to improve the reliability of complex tasks. Traditional fault detection methods rely on prior knowledge or models and have been widely applied. However, when dealing with complex systems, multi-source and high-dimensional data, fault detection techniques based on intelligent algorithms are showing their advantages. Firstly, based on the characteristics of deep-sea environment and the navigation scheme of underwater vehicles, the traditional methods and the detection methods based on intelligent algorithms are introduced. Then, the characteristics of different detection methods and their applications in underwater environments are analyzed based on case studies, and the research progress of fault detection methods is summarized. Finally, the research direction of fault detection methods for future deep-sea submersible integrated navigation systems is discussed.

Accurate identification of pedestrian motion modals can aid gait detection. For the problem of inaccurate gait detection under pedestrian continuous motion modals, a multi-conditional constraint gait detection algorithm based on modal identification is proposed in this paper. Firstly, based on the motion characteristics of pedestrians and the pose characteristics of handheld terminals, finite state machine idea is introduced to construct lightweight classifiers to identify three motion states of pedestrians(stationary, walking and running) and three usage modes of terminals(flat end, telephone and swing arm), in order to improve the adaptability of gait detection. Then, the gait detection accuracy is improved by extreme value constraint, periodicity constraint and peak/valley matching mechanism. Finally, the pseudo-peaks/valleys are eliminated by the adjacent peak/valley substitution mechanism to achieve more accurate gait detection. The experiment results show that the algorithm has a minimum gait detection accuracy of 98.89% in a single-motion modal and an overall detection accuracy of 100% in continuous complex modals, which has strong accuracy and robustness.

The integrated navigation system is a typical nonlinear system. The model nonlinearity and noise uncertainty will affect the estimation accuracy of the system. Aiming at the problem that the filtering accuracy of traditional unscented Kalman filter decreases in the case of non-Gaussian noise, an improved adaptive maximum correntropy unscented Kalman filter algorithm is proposed. This method estimates the measurement noise at the current moment according to the filtering innovation, and determines the change interval of the kernel width by using the change degree of the estimated noise relative to the historical noise. The kernel width is adaptively adjusted according to the iterative error change of the maximum correntropy algorithm, which improves the convergence speed of the algorithm and the processing ability of non-Gaussian noise. Based on the non-Gaussian noise environment, the SINS/GNSS integrated navigation simulation experiment is built and the actual car experiment is carried out. Compared with the traditional unscented Kalman filter, the results show that the east position error and north position error obtained by the proposed algorithm are 2.11 m and 1.85 m under the condition of non-Gaussian noise. The filtering performance is obviously better than the traditional UKF, which improves the accuracy of the integrated navigation solution.

The gravity benchmarks are scarce in Qinghai Province and cannot meet the needs of various industries. The methods and experiences of precision control are introduced through the construction of a new basic network. Firstly, 18 gravity observation piers are constructed through exploration, selection, and burial. Then, four CG-5 and two CG-6 high-precision relative gravimeters are used for gravity measurement with reference points, basic points, and introduction-points, with a maximum error of 7.2 μGal in the measurement. Finally, 18 gravity point results are obtained through grid value conversion, solid tide correction, instrument height correction, atmospheric pressure correction, zero drift correction, and gravity segment difference calculation of the observation data. After adjustment calculation, the unit weight mean square error is 12.2 μGal, with a mean square error of 4.3 μGal for the average gravity point value, all accuracy indicators are better than the standard requirements.

As an important part of aerial photoelectric pod, photoelectric platform is widely used in UAV, seeker and other fields. In order to compensate friction, angular velocity disturbance and other unknown disturbances and improve the system stabilizing accuracy, a composite control method based on fuzzy tuning rules and improved disturbance observer(IDOB) is proposed. In this method, fuzzy rules are used to adjust PID parameters online, and an improved disturbance observer based on angular velocity is used to observe the external disturbance in the servo control system, which is equivalent to the control quantity for feedforward compensation. The prototype swing experiment results show that the RMS value of the proposed compound control method is 12.24% less than that of the fuzzy PID control under the condition of low frequency disturbance, which effectively improves the line of sight stabilizing accuracy and robustness under low frequency disturbance motion. The composite control method has better stability, and has reference significance for engineering practice.

The coherent population trapping(CPT) atomic magnetometer can measure magnetic field scalar with high-precision, but it is difficult to make high-precision measurements of the magnetic field vector. In order to achieve the measuring ability of geomagnetic field vector, a method based on CPT effect for measuring the magnetic field vector using dual linearly polarized light path is proposed in this paper. In this method, a VCSEL light source is modulated by a 3.4 GHz microwave module to produce two linearly polarized pump lights that satisfy the CPT effect. The two linearly polarized pump lights, which propagation directions are perpendicular to each other, are injected into ⁸⁷Rb atomic vapor cell through a light path structure. The polarization direction of the two linearly polarized pump lights is adjusted. The polarization direction corresponding to the extremum of the CPT signal peak is measured. The angular position of the magnetic field in the plane perpendicular to the light propagation direction is subsequently determined. The amplitude and direction of the magnetic field vector are obtained from measurements of the two lights. The experiment results show that angle error is about 1° under the condition of a total magnetic field of 2133 nT. The feasibility of the proposed method for measuring magnetic field vector is proved. This method has the advantages of not being affected by signal amplitude fluctuations, high theoretical measurement accuracy, and no need for tri-axis coil assistance, and it can simplify the magnetic field measurement system, which is of great significance in satellite magnetic measurement.

In inertial navigation system, the current signal output by the accelerometer needs to be converted into the output pulse through the analog-to-digital conversion circuit. Aiming at the problem for nonlinear distortion due to the unsatisfactory device in output of I/F analog-to-digital conversion circuit in inertial navigation system, a two-stage compensation algorithm is proposed in this paper. In this algorithm, temperature compensation model and linearity compensation model are established to compensate the linearity of the circuit, the error caused by temperature change is compensated, and the problem that the hardware only compensates the first-order error is solved, which makes the analog-to-digital conversion circuit maintain high-precision in large range environment even if the device parameters are unsatisfactory. The algorithm is applied to the practical circuits, and the experiment results show that this algorithm can reduce the nonlinear index of I/F analog-to-digital conversion circuit from 7.20×10^-4 to 4.93×10^-6, and effectively decrease the nonlinear distortion caused by the unsatisfactory device.

With the increasing demands for assembly accuracy and automation of large cabin products such as aircraft and rocket in the aerospace field, the assembly modes of large cabin are moving from the original manual operation mode of “visual shouting and manual pushing” to the direction of automation, digitalization and intelligence. In this paper, to address this issue, an automatic docking system based on machine vision is designed, and a method for measuring the relative pose of large cabin based on two sets of monocular vision is proposed. The system uses two sets of monocular vision to independently measure the position and attitude of the target on the surface of the module, and combines the pre-calibrated position relationship between the target and the module, and integrates the two sets of monocular vision measurement results to calculate the final docking attitude of the module. The experiment results show that the measurement accuracy of the system is better than 0.04 mm, and the docking accuracy is better than 0.3 mm, which has strong application value.

Aiming at the problem of difficult description of the complex structure and the movement position of the internal frames of the inertial platform system during testing, the visualization of the digital model of the inertial platform and a method of the model driven are proposed. A three-dimensional visualization system of inertial platform for online testing and offline teaching is developed by using Qt/OpenGL. The inertial platform is modeled by a CAD modeling software, and the interface between OpenGL and the CAD software, as well as the interface between digital prototype and actual platform, are determined. The verification based on online/offline platform tests shows that the developed system realizes the real-time tracking of the digital prototype to the actual movement of the platform frames. The structure and positions of the frames during online/offline testing are displayed clearly. The developed system effectively improves the onsite communication during testing and the training among platform testing.

The single-pixel camera is not yet mature in terms of integration and engineering, and most research only focuses on building and testing in laboratory environments. However, in engineering applications, the volume, weight, and testing environment of the prototype are all factors that affect the quality of single-pixel imaging. Therefore, under the constraints of volume, weight, and testing environment, a single pixel camera in the near infrared band is tested in complex environments such as natural light sources and infrared lights using an efficient encoding and decoding algorithm integrated with C++ language. The integrated single-pixel camera can achieve 2.5 frames per second @256×256 pixels with an angular resolution of 7.3 mrad at a 6.25% sampling rate. As the exposure time increases, the image signal-to-noise ratio significantly rises. After processing with a deep learning denoising network, the quality of single-frame imaging still has room for improvement. Additionally, the response characteristics of the integrated prototype to vibration are also simulated and studied, and the adaptability of the prototype in general transportation environments is discussed. This study has a promoting effect on improving the technological maturity of single-pixel cameras, and provides a solution for low-cost infrared imaging.

A new operation and maintenance technology based on digital twin for strapdown inertial navigation system has been proposed to address the complex data analysis, inaccurate fault localization, and difficulty in managing the lifecycle of the production process. By combining digital twin technology with strapdown inertial navigation system, it offers innovative solutions for the intricate design, production, and lifecycle maintenance of strapdown inertial navigation system. By designing a digital twin system based on the full element visualization of the strapdown inertial navigation system, the equipment is managed for lifecycle data. Results from production show that this technology can replicate physical entities in a virtual space and achieve real-time interaction between physical and digital spaces throughout the lifecycle of strapdown inertial navigation system by using historical data, real-time data, technical knowledge, processing algorithms, functional models, thus establish a fault database and enable digital health management.

In the process of feature extraction and matching in aircraft visual navigation, the imaging methods of heterogeneous images are different, and the feature information varies greatly. Due to different standards, it is difficult to achieve accurate matching for various types of image information. Considering the above issues, a heterogeneous image matching algorithm based on feature fusion is proposed in this paper, which can achieve high-precision matching of heterogeneous images. Firstly, an encoding network based on residual networks is designed to enhance the ability to extract deep level features of images. Then, dense connections are introduced in the decoding network to preserve intermediate layer features, enabling the fusion results to take into account information at all levels of the image. Besides, a fusion strategy including attention mechanism is designed to optimize the information proportion of the source image in the fusion features and improve the quality of the fusion features. Finally, the fused features are applied to image matching. The experiment results show that the proposed algorithm exhibits advantages in feature fusion evaluation indicators compared to common methods. In the experiment of heterogeneous image matching, the matching accuracy of the proposed algorithm is 96.71% and the matching accuracy under rotation is 89.40%, which reflects high accuracy and strong robustness performance.