Over the last few years, at ByteDance, our compute infrastructure scale has been expanding significantly due to expedited business growth. In this journey, to meet hyper-scale growth, some business groups resorted to managing their own compute infrastructure stack running different scheduling systems such as Kubernetes, YARN which created two major pain points: the increasing resource fragmentation across different business groups and the inadequate resource elasticity between workloads of different business priorities. Isolation across different business groups (and their compute infrastructure management) leads to inefficient compute resource utilization and prevents us from serving the business growth needs in the long run.

The evolving fuzzy technique is a recent development in the soft-computing field that has shown some promising results in applications such as control, classification, and short-term prediction. However, evolving fuzzy techniques still have challenges in terms of high-speed processing of cluster/rule generation, especially in long-term prediction applications due to the broader distribution of the input space. These factors can lead to problems such as overfitting in optimization and high computational costs, which could limit their applications in real-time monitoring. In this article, an adaptive evolving fuzzy (AEF) technique consisting of two novel aspects is developed to tackle these problems. First, an error-assessment method is suggested to monitor the trend of the cumulative training errors and to control the fuzzy cluster evolving process. Second, an adaptive particle filter algorithm is proposed to optimize the fuzzy clusters in order to enhance incremental learning and improve modeling efficiency. The effectiveness of the proposed AEF predictor is verified by simulation tests; it is also implemented for battery remaining useful life forecasting. Test results have shown that the proposed AEF technique can effectively capture the system's dynamic characteristics with fewer rules and can provide more flexibility in fuzzy modeling.

Human society is rapidly transforming to the digital society, the new technologies drive digital and intelligent transformation in all industries. These technologies promise cost savings, efficiency gains, and new value. At the same time, we see growing challenges relating to cyber security and privacy protection and functional safety etc. As a leading ICT product and service provider, Huawei has been committed to providing customers with high-quality and user-friendly products and services. Two years ago, Huawei initiated the Transformation Program for Software Engineering Capability Enhancement, to improve company-wide software engineering capabilities, improve the trustworthiness of both processes and results, and to provide trustworthy and quality products. This talk will share the practices, progress, and challenges of the transformation.

A universal approximation of multi-input multi-output (MIMO) fuzzy systems is proposed in this article based on a fuzzy relation matrix (FRM) method. The fuzzy reasoning operation in FRM is realized by the semitensor product (STP) technique. The theoretical proof is provided to estimate approximation accuracy of the FRM models for MIMO systems. Its effectiveness is verified by simulation. Simulation results show that the proposed fuzzy design technology is efficient for fuzzy systems modeling, and the general universal approximation theory can be extended to the MIMO fuzzy systems. The proposed uniform matrix expression can be used to design fuzzy systems with FRM models and to approximate nonlinear functions with required accuracy.

A new fuzzy formulation technique based on semitensor product of matrices is proposed in this paper to construct multivariable fuzzy logic systems. A fuzzy relation matrix (FRM) model is suggested for direct modeling and indirect identification of the fuzzy matrix expression using the observed input-output data of fuzzy dynamical process. The center points of fuzzy sets for every output variable are optimized by using least squares estimation. The effectiveness of the proposed fuzzy formulation technology is verified by simulations. Test results show that the proposed formulation technique is an effective design method to construct and optimize fuzzy relation structure matrices for multivariable fuzzy systems. It can be used to design FRM models for multi-input multi-output systems.

This article has been retracted: please see Elsevier Policy on Article Withdrawal (https://www.elsevier.com/about/our-business/policies/article-withdrawals). This article has been retracted at the request of .

In many system state forecasting applications, the prediction is performed based on multiple datasets, each corresponding to a distinct system condition. The traditional methods dealing with multiple datasets (e.g., vector autoregressive moving average models and neural networks) have some shortcomings, such as limited modeling capability and opaque reasoning operations. To tackle these problems, a novel fuzzy neural network (FNN) is proposed in this paper to effectively extract information from multiple datasets, so as to improve forecasting accuracy. The proposed predictor consists of both autoregressive (AR) nodes modeling and nonlinear nodes modeling; AR models/nodes are used to capture the linear correlation of the datasets, and the nonlinear correlation of the datasets are modeled with nonlinear neuron nodes. A novel particle swarm technique [i.e., Laplace particle swarm (LPS) method] is proposed to facilitate parameters estimation of the predictor and improve modeling accuracy. The effectiveness of the developed FNN predictor and the associated LPS method is verified by a series of tests related to Mackey-Glass data forecast, exchange rate data prediction, and gear system prognosis. Test results show that the developed FNN predictor and the LPS method can capture the dynamics of multiple datasets effectively and track system characteristics accurately.

A smart monitor is developed in this paper for vibration measurement and signal analysis. The 3D vibration signals can be measured in either an analog or a digital form. This reprogrammable monitor can conduct fault detect in shaft-related systems such as rotor imbalance, misalignment, and bearing defects. An energy spectrum technique is proposed and implemented for bearing fault detection: Firstly bearing resonance signatures are demodulated by wavelet decomposition, and then the resulting wavelet energy functions are integrated to enhance feature characteristics. A correlation spectrum is employed to highlight bearing fault characteristic frequencies. The effectiveness of the proposed techniques is verified by experimental tests corresponding to different bearing conditions. Test results show that the suggested wavelet energy spectrum approach is a robust bearing fault detection technique especially for non-stationary feature extraction and analysis.

The commonly used genetic algorithms (GAs) have some shortcomings in applications such as lengthy computations and slow convergence. A novel enhanced genetic algorithm, EGA, technique is developed in this paper to overcome these problems to enhance the efficiency in system training and optimization. The proposed EGA technique involves two approaches: a) a novel group-based branch crossover operator is suggested to thoroughly explore local space and to speed convergence, and b) an enhanced MPT (Makinen-Periaux-Toivanen) mutation operator is proposed to promote global search capability. The effectiveness of the developed EGA is verified by simulations using benchmark test problems. Test results show that the EGA technique can improve the classical GA methods with respect to convergence speed and global search capability.

This paper investigates the problem of global output feedback tracking control of flexible joint robots. Despite the fact that only link position and actuator position are available from measurements, the proposed controller ensures that the link position globally tracks the desired trajectory while keeping all the remaining signals bounded. The controller development uses a partial state-feedback linearization technique combined with the integrator backstepping control design method whereas a filter and an observer are utilized to remove the requirement of link and actuator velocity measurements. Partial state-feedback linearization of robot dynamics is performed by factoring the manipulator mass matrix into a quadratic form involving an integrable root matrix. The applicability of the proposed general design methodology is illustrated by an example of flexible joint planar robots. Numerical results for a two-link flexible joint planar robot are also provided.

A reliable monitoring system is critically needed in a wide range of industries to detect the occurrence of a fault to prevent machinery performance degradation, malfunction, and sudden failure. In this paper, a new intelligent system, extended neurofuzzy (ENF) scheme, is proposed for real-time machinery condition monitoring. The monitoring reliability is improved by integrating the predicted machinery condition to fault diagnosis. The ENF scheme can perform both classification and prediction operations. The ENF classifier integrates the merits of several signal processing techniques for a more positive assessment of the machinery condition. The ENF predictor forecasts the machinery condition propagation trends. An interscheme training technique is proposed to improve the ENF system's adaptive capability to accommodate different operation conditions. The viability of this new monitoring system has been verified by experimental tests. Test results have shown that the developed ENF system is a robust condition monitoring tool that has good adaptive capabilities to accommodate different machinery conditions.

In the above titled paper (ibid., vol. 16, no. 1, pp. 110-122, Jan 08), the equation (14) and the text which follows should read as presented here.

A reliable and online predictor is very useful to a wide array of industries to forecast the behavior of time-varying dynamic systems. In this paper, an evolving fuzzy system (EFS) is developed for system state forecasting. An evolving clustering algorithm is proposed for cluster generation. Clusters are established and modified based on constraint criteria of mapping consistence and compatible measurement. A novel recursive Levenberg-Marquardt (R-LM) method is proposed for online training of nonlinear EFS parameters. The viability of the developed EFS predictor is evaluated based on both simulation from benchmark data and real-time tests corresponding to machinery condition monitoring and material property testing. Test results show that the developed EFS predictor is an effective and accurate forecasting tool. It can capture the system's dynamic behavior quickly and track the system's characteristics accurately. The proposed clustering algorithm is an effective structure identification method. The recursive training technique is computationally efficient, and can effectively improve reasoning convergence.

The detection of the onset of damage in gear systems is of great importance to industry. In this paper, a new neuro-fuzzy diagnostic system is developed, whereby the strengths of three robust signal processing techniques are integrated. The adopted techniques are: the continuous wavelet transform (amplitude) and beta kurtosis based on the overall residual signal, and the phase modulation by employing the signal average. Three reference functions are proposed as post-processing techniques to enhance the feature characteristics in a way that increases the accuracy of fault detection. Monitoring indexes are derived to facilitate the automatic diagnoses. A constrained-gradient-reliability algorithm is developed to train the fuzzy membership function parameters and rule weights, while the required fuzzy completeness is retained. The system output is set to different monitoring levels by using an optimization procedure to facilitate the decision-making process. The test results demonstrate that the novel neuro-fuzzy system, because of its adaptability and robustness, significantly improves the diagnostic accuracy. It outperforms other related classifiers, such as those based on fuzzy logic and neuro-fuzzy schemes, which adopt different types of rule weights and employ different training algorithms.