Supervised training of spiking neural networks for robust deployment on mixed-signal neuromorphic processors

Mixed-signal analog/digital circuits emulate spiking neurons and synapses with extremely high energy efficiency, an approach known as “neuromorphic engineering”. However, analog circuits are sensitive to process-induced variation among transistors in a chip (“device mismatch”). For neuromorphic implementation of Spiking Neural Networks (SNNs), mismatch causes parameter variation between identically-configured neurons and synapses. Each chip exhibits a different distribution of neural parameters, causing deployed networks to respond differently between chips. Current solutions to mitigate mismatch based on per-chip calibration or on-chip learning entail increased design complexity, area and cost, making deployment of neuromorphic devices expensive and difficult. Here we present a supervised learning approach that produces SNNs with high robustness to mismatch and other common sources of noise. Our method trains SNNs to perform temporal classification tasks by mimicking a pre-trained dynamical system, using a local learning rule from non-linear control theory. We demonstrate our method on two tasks requiring temporal memory, and measure the robustness of our approach to several forms of noise and mismatch. We show that our approach is more robust than common alternatives for training SNNs. Our method provides robust deployment of pre-trained networks on mixed-signal neuromorphic hardware, without requiring per-device training or calibration.

Circuit Parameter Range of Photovoltaic System to Correctly Use the MPP Linear Model of Photovoltaic Cell

The real-time linearization of a photovoltaic (PV) cell has been implemented well by the proposition of two maximum power point (MPP) linear models (MPP Thevenin cell model and MPP Norton cell model). However, there is no work to specially analyze the circuit parameter range (CPR) to correctly use them, which seriously impedes the development of the linear control theory involving them. To deal with this problem, in this paper, PV systems with three usual outputs are analyzed and the expressions of their CPR are proposed under ideal conditions. Meanwhile, these expressions are improved to match the practical application. They disclose the relationships between load (or bus voltage) and model parameters of the MPP Thevenin cell model (MPP-TCM) when the MPP of PV system always exists. They also reveal the constraints of load (or bus voltage) when the MPP-TCM is always available. Finally, by some simulation experiments, the accuracy of the expressions of the CPR is verified, the regular patterns of the CPR changing with weather are disclosed, and the comparison of the CPR for different PV systems are made. In this work, the relationships between MPP-TCM and circuit parameters are successfully found, disclosing the constraints among parameters when the MPP-TCM is used to implement the overall linearization of a PV system.

Feedback Control for a Drone with a Suspended Load via Hierarchical Linearization

As a possible extension of a drone application, transportation of a cable-suspended load is expected. The model of a drone with a suspended load is a nonlinear underactuated system that is known to be difficult to analyze and control. This paper applies the linearization method, known as hierarchical linearization, to the system. We observed that, via the hierarchical linearization scheme, the system can be linearized exactly and the controller can be designed simultaneously. There are two features of this approach. First, the controller exactly considers the system nonlinearity, and the feedback controller is based on the linear control theory. Second, it is possible to derive the analytical solution of the closed-loop system. We have demonstrated these features via numerical simulations.

Optimization of a control law to synchronize first-order dynamical systems on Riemannian manifolds by a transverse component

<p style='text-indent:20px;'>The present paper builds on the previous contribution by the second author, S. Fiori, <i>Synchronization of first-order autonomous oscillators on Riemannian manifolds</i>, Discrete and Continuous Dynamical Systems – Series B, Vol. 24, No. 4, pp. 1725 – 1741, April 2019. The aim of the present paper is to optimize a previously-developed control law to achieve synchronization of first-order non-linear oscillators whose state evolves on a Riemannian manifold. The optimization of such control law has been achieved by introducing a transverse control field, which guarantees reduced control effort without affecting the synchronization speed of the oscillators. The developed non-linear control theory has been analyzed from a theoretical point of view as well as through a comprehensive series of numerical experiments.</p>

Application of Mobile Robots to Linear Control Theory Education

The knowledge of control engineering for mechanical engineers seems to become more important with the continuous development of automated technologies. To cultivate this knowledge, many experimental devices have been proposed and used. Devices with direct current (DC) motors are widely used because the DC motors can be controlled with sufficient accuracy based on the classical linear control theory. Mobile robots are used as educational platforms attracting the attention of students in various problem-based learning subjects. However, they have been hardly used to teach linear control theory because of the nonlinearity. This paper shows an experimental curriculum to learn control theory using a mobile robot instead of a motor. Although the model of the mobile robot is nonlinear, a strict linearization method makes it possible to adjust the control gains using the linear control theory. By applying the method, the characteristics of linear control systems are explicitly observed in the traveling paths of the mobile robot, so an experimental curriculum to learn the basic linear control theory can be realized using an inexpensive mobile robot. The proposed experimental curriculum was carried out in a class of a mechanical engineering course, and its results are discussed in this paper.

Synthesis of robust control algorithms for linear interval dynamic systems in the electric power industry

When studying the objects of management in the field of electric power, we meet with various inaccuracies in determining their parameters. One of the methods of dealing with uncertainties is the use of various estimates of the parameters of the control object. modern science has developed various methods for assessing the uncertain parameters of the control object in the electric power industry. parameter uncertainty occurs when the set of parameters of the control object is more than one point. If this set is defined using probabilistic characteristics, then this is the so-called probabilistic uncertainty of the object parameters. If the boundaries of the intervals in which they are enclosed are known for the object parameters, then such uncertainty is called interval uncertainty. if the object parameters are set using the membership function, then the theory of fuzzy logic is used. Interval determination of parameters of the control object is used when working with values for which only the boundaries of the intervals within which their values are enclosed are known. The interval approach in the description of object parameters is used to account for rounding and errors that occur during calculations on a computer and is a strong method in the representation of objects with uncertain parameters, which are very common in the electric power industry. The reason why interval systems are used is the incompleteness of information about the control object, errors in measuring the parameters of the object, linearization errors, and so on. Various problems of the classical theory of automatic control allow us to replace the concentrated parameters with their interval analogues. Many interval problems are adequate for practical applications. The synthesis of linear quadratic regulators refers to the classical method, which allows us to obtain regulators that minimize the integral quality criterion with respect to the resource of regulated and regulatory quantities. In this paper, we study the possibility of synthesizing linear quadratic regulators for electric power facilities specified in intervals. This application of two well-known methods makes it possible to work with nonlinear objects using the classical linear control theory.

Manufacturing and Performance of an Economical 1-D Shake Table

The researchers and engineers encountered many problems to precisely replicate earthquake waves. Earthquakes are one of the nature's worst catastrophes and are still unpredictable. Statistical research has shown that the earthquakes have increased in frequency in recent years and have become a major concern for the world especially for those countries which are located on the fault lines such as Japan, Bangladesh and Pakistan. So, it was imperative to device a mechanism to check earthquake response and apply some necessary mitigations for the safety of humanity. After many years of research an indispensable testing apparatus was designed named as Shake Table. This apparatus is extensively used in earthquake research centers globally because it is the best available apparatus to replicate the earthquakes imposed dynamic effects on structures. A uni-axial shaking table was designed, manufactured and installed in University of Engineering & Technology Taxila, Pakistan which is operated on 3 HP servo motor coupled with encoder, motion controller and supported on HSB mechanical linear drive. The system was assembled in a simple way with care to endure sufficient replication of given (recorded) motion by shake table system. This paper focuses on the designing, manufacturing and performance of an economical analytical model of 1-D shake table incorporating conjunction of structural dynamics and linear control theory.

Passive Current Control Design for MMC in HVDC Systems through Energy Reshaping

The complexity of the internal dynamics of a modular multi-level converter (MMC) has raised severe issues for designing corresponding controllers. The existing MMC cascaded control strategies, based on classical linear control theory, require a relatively complex structure to achieve control objectives and the parameter tuning processes during the corresponding controller design are normally difficult to solve for the highly non-linear systems with highly coupled states in MMC. On account of this, advanced controllers are required for the regulation tasks of MMC. Passivity is introduced into the MMC control system by the passive control (PC) proposed in this paper. PC can provide an extra damping effect to help save energy through utilizing passivity in the system. A controllable de-coupled form is achieved by passivation of the output calculation. Hence, well-tuned controllers can be designed and employed to effectively regulate the output current and inner differential currents of the MMC under system operating point variation. Simulation results yield numerical data that show significantly improved steady-state and transient-state performances with greatly reduced control costs.