The mass spectrometer filament current sensor reflects the health status of the filament by detecting real-time variation in filament current, providing an ideal means to monitor the safe use of the filament. However, the complex and changing environment inside and outside the sensor often affects the sensor, so it is important to perform fault diagnosis and calibration of the sensor. To address the problems of low fault identification accuracy and high calibration error of a single sensor due to insufficient fault samples, a CNN-LSTM-based fault diagnosis and adaptive multi-channel fusion calibration of filament current sensors for mass spectrometers is proposed. First, two valid sensor fault datasets are generated by numerical simulation of on-orbit historical data collected by a spacecraft through the CVAE model. Secondly, the fault identification of the dataset is performed by the CNN-LSTM model, and the designed weighted fusion loss function leads to the improved accuracy of the model. Finally, an adaptive multi-channel calibration model is constructed, and a method for adaptively selecting calibration channels based on the characteristics of the fault samples themselves is proposed, which significantly reduces the calibration error compared with a single calibration model. The experimental results show that the method can improve the fault identification rate and reduce the calibration fault error compared with other fault diagnosis and calibration methods. It provides an effective sensor fault solution for mass spectrometer filament current sensors.

Safety and stability are critical in manned space missions, requiring an environmental control and life support system (ECLSS) of spacecraft to operate reliably. This study analyzed the time-series characteristics of telemetry data, including total pressure, temperature, and humidity, to predict the ECLSS’s operational state. Existing algorithms for time-series forecasting, including ARIMA, LSTM, TCN, and NBEATS, often struggle with long-sequence forecasting and discrepancies in data distribution, which hinder their ability to deliver accurate predictions. To address these challenges, this study introduces a two-stage normalization method, mean instance normalization (MeanIN), designed to adjust input data distributions and restore output data distributions, thereby significantly enhancing predictive performance. Experimental evaluations on ECLSS telemetry data demonstrate that MeanIN consistently improves model accuracy, with the informer model achieving superior results in long-sequence forecasting tasks. These results underscore the efficacy of MeanIN and its potential to support critical applications in anomaly detection and predictive analysis for spacecraft telemetry data.

Due to the disadvantages such as complex calculation, low accuracy of estimation, and being non real time in present methods, a new real-time algorithm is developed for on-orbit magnetometer bias determination of micro-satellites without attitude knowledge in this paper. This method uses the differential value approach. It avoids the impact of quartic nature and uses the iterative method to satisfy real-time applications. Simulation results indicate that the new real-time algorithm is more accurate compared with other methods, which are also tested by an experiment system using real noise data. With the new real-time algorithm, a magnetometer calibration can be taken on-orbit and will reduce the demand for computing power effectively

High integration of multi-functional instruments raises a critical issue in temperature control that is challenging due to its spatial–temporal complexity. This paper presents a multi-input multi-output (MIMO) self-tuning temperature sensing and control system for efficiently modulating the temperature environment within a multi-module instrument. The smart system ensures that the internal temperature of the instrument converges to a target without the need of a system model, thus making the control robust. The system consists of a fully-connected proportional–integral–derivative (PID) neural network (FCPIDNN) and an on-line self-tuning module. The experimental results show that the presented system can effectively control the internal temperature under various mission scenarios, in particular, it is able to self-reconfigure upon actuator failure. The system provides a new scheme for a complex and time-variant MIMO control system which can be widely applied for the distributed measurement and control of the environment in instruments, integration electronics, and house constructions.

Multi-physical field sensing and machine learning have drawn great attention in various fields such as sensor networks, robotics, energy devices, smart buildings, intelligent system and so on. In this paper, we present a novel efficient method for thermal and energy management based on bimodal airflow-temperature sensing and reinforcement learning, which expedites an exploration process by self-learning and adjusts action policy only through actuators interacting with the environment, being free of the controlled object model and priori experiences. In general, training of reinforcement learning requires a large amount of data iterations, which takes a long time and is not suitable for real-time control. Here, we propose an approach to speed up the learning process by indicating the action adjustment direction. We adopt tailor-designed bimodal sensors to simultaneously detect airflow and temperature field, which provides comprehensive information for reinforcement learning. The proposed thermal and energy management incorporates bimodal parametric sensing with an improved actor-critic algorithm to realize self-learning control. Experiments of thermal and energy management in a multi-module integrated system validate the effectiveness of the proposed methodology, which demonstrate high efficiency, fast response, and good robustness in various control scenarios. The proposed methodology can be widely applied to thermal and energy management of diverse integrated systems.

Artificial Neural Networks (ANNs), including Deep Neural Networks (DNNs), have become the state-of-the-art methods in machine learning and achieved amazing success in speech recognition, visual object recognition, and many other domains. There are several hardware platforms for developing accelerated implementation of ANN models. Since Field Programmable Gate Array (FPGA) architectures are flexible and can provide high performance per watt of power consumption, they have drawn a number of applications from scientists. In this paper, we propose a FPGA-based, granularity-variable neuromorphic processor (FBGVNP). The traits of FBGVNP can be summarized as granularity variability, scalability, integrated computing, and addressing ability: first, the number of neurons is variable rather than constant in one core; second, the multi-core network scale can be extended in various forms; third, the neuron addressing and computing processes are executed simultaneously. These make the processor more flexible and better suited for different applications. Moreover, a neural network-based controller is mapped to FBGVNP and applied in a multi-input, multi-output, (MIMO) real-time, temperature-sensing and control system. Experiments validate the effectiveness of the neuromorphic processor. The FBGVNP provides a new scheme for building ANNs, which is flexible, highly energy-efficient, and can be applied in many areas.