The problem of decentralized event-triggered control for a class of network-based state-dependent uncertain systems subject to network transmission delay and deception attacks is considered in this article. To reduce network load, a novel decentralized adaptive event-triggered scheme (AETS) is developed to transmit necessary sampled signals. During network transmission, a more practical deception attack phenomenon is considered, where the attack behaviors in different channels are governed by independent Bernoulli processes. Moreover, a set of improved data buffers are applied in the controller side to organize the decentralized triggered data and alleviate the impact of network transmission delay, such that the transmitted data can be utilized timely. Then, an integrated closed-loop system with state-dependent uncertainties is constructed by taking the AETS, deception attacks and data buffers into account. Sufficient conditions that guarantee the mean-square exponential stability of the closed-loop system are presented by employing the Lyapunov functional method, and the design criterion of the controller gain is given by an exact expression. Finally, the proposed method is applied to the control of electronic circuits to verify its practicability and effectiveness.

In this paper, the event-triggered robust filtering problem is investigated for a class of network-based state-dependent uncertain systems in the sense of finite-time boundedness, considering network transmission delay. By resorting to a generalized performance index, the H∞ and L₂-L∞ filtering problems can be solved in a unified framework. An event-triggered scheme is also utilized to sift out the sampled signals so that both the communication resources and the network bandwidth can be saved as much as possible. By means of an appropriate Lyapunov-Krasovskii functional and the delay system approach, sufficient conditions guaranteeing the filtering error system being finite-time bounded with extended dissipativity are deduced. Based on the convex optimization, an explicit expression of the desired finite-time robust filter is derived. Finally, two numerical examples are provided to illustrate the effectiveness of the developed methods.

A data-driven intelligent fault diagnosis algorithm is designed in this paper for heavy-duty industrial-robots (I-Rs), which aims accurately detecting and identifying joint actuator faults that may occur during the operation of industrial heavy-duty robot arms. Considering the fact that faulty samples of industrial robots are difficult to access, a simulation model-based strategy is adopted in this paper. With the help of Euler-Lagrange method, dynamics of heavy-duty industrial robots are established with considering the joint flexibility. By injecting fault in different joint actuators, unbalanced normal and faulty samples are obtained, based on which intelligent diagnosis model is constructed. Subsequently, a composite neural network model, LSTM-CNN, is proposed, which combines the merits of long short-term memory network (LSTM) and convolutional neural network (CNN). The constructed LSTM-CNN model is then trained and validated using generated data to achieve fault diagnosis and identification of actuators of heavy industrial robots. Finally, the constructed intelligent fault diagnosis model is experimentally validated, and result analysis demonstrates that the proposed algorithm shows superior accuracy and precision in diagnosing single joint and multiple joints faults.

Glioblastoma (GBM) is the most common and aggressive malignant brain tumor in adults and is poorly controlled. Previous studies have shown that both macrophages and angiogenesis play significant roles in GBM progression, and co-targeting of CSF1R and VEGFR is likely to be an effective strategy for GBM treatment. Therefore, this study developed a novel and selective inhibitor of CSF1R and VEGFR, SYHA1813, possessing potent antitumor activity against GBM. SYHA1813 inhibited VEGFR and CSF1R kinase activities with high potency and selectivity and thus blocked the cell viability of HUVECs and macrophages and exhibited anti-angiogenetic effects both in vitro and in vivo. SYHA1813 also displayed potent in vivo antitumor activity against GBM in immune-competent and immune-deficient mouse models, including temozolomide (TMZ) insensitive tumors. Notably, SYHA1813 could penetrate the blood–brain barrier (BBB) and prolong the survival time of mice bearing intracranial GBM xenografts. Moreover, SYHA1813 treatment resulted in a synergistic antitumor efficacy in combination with the PD-1 antibody. As a clinical proof of concept, SYHA1813 achieved confirmed responses in patients with recurrent GBM in an ongoing first-in-human phase I trial. The data of this study support the rationale for an ongoing phase I clinical study (ChiCTR2100045380).

Many critical applications necessitate indoor positioning, and the implementation of indoor positioning cannot be separated from various wireless signals. Among these, 5G-based indoor positioning is gaining increasing attention due to the wide coverage of 5G base stations and the stability of 5G signals. However, current positioning algorithms based on 5G signals require specialized measuring tools or information from telecommunication operators. Therefore, this paper proposes an indoor positioning algorithm, called 5G1M, that uses only a single 5G module and several commercial base stations already built outdoors. Besides these, 5G1M does not need any further assistance. We design 5G1M based on fingerprint positioning, which is less dependent on the environment and devices. First, 5G1M incorporates a Siamese network model with Ghost modules and squeeze-and-excitation blocks to reduce reliance on the scale of the fingerprint database and computational power while maintaining accuracy. Then, transfer learning is used to adapt to long-term environmental changes, and a trajectory-fitting method is employed to mitigate instantaneous signal fluctuations. The experimental result shows that in such a simplified positioning platform, 5G1M achieves a positioning error of 2.53 meters, improving accuracy by 29.7% compared to existing algorithms, and it also demonstrates faster convergence with fewer model parameters.