RU2546993C1 - Method to diagnose technical condition of electric drive by evaluation of dynamics of its electrical parameters - Google Patents

Abstract

FIELD: electricity.

SUBSTANCE: invention is related to the area of technical condition diagnostics for electrical wires, for example, for electric drives of rolling mills in metallurgical production on the basis of parameters analysis of current, voltage, speed and control setting with use of recurrent artificial neuron network. The concept of the invention is as follows: parameters of current, voltage, speed and control setting are measured with a certain time interval, and then parameters are converted to digital form and transferred to PC for processing. Before operation of the drive the software-based recurrent artificial neuron network trained for the specific electric drive reproduces dynamics of the drive parameters, thereupon results of the network model dynamics are compared with real dynamics of the electric drive. In faulty electric drive dynamics of its parameters deviate from the model dynamics and discrepancy function is calculated. By nature of discrepancy function evaluation of technical condition and forecast of the drive resource is made.

EFFECT: improved accuracy and reliability for diagnosing of emergency conditions of the electric drive at operating equipment at early stage and non-observed stage thus preventing sudden emergency shutdown of the electric drive and reducing repair costs.

2 dwg

Description

The invention relates to the field of diagnostics of the technical condition of an electric drive by assessing the dynamics of its parameters and can be used for diagnostics without decommissioning electric drives of various types operating in difficult conditions where constant access to personnel is difficult or impossible, as well as drives of critical mechanisms where emergency situations are required in the early stages, without causing serious damage.

There is a method for diagnosing the technical condition of an electric motor according to its electrical parameters, described in RU 2425391 C1, G01R 31/34, METHOD FOR DIAGNOSTICS OF THE TECHNICAL CONDITION OF THE ELECTRIC MOTOR BY ITS ELECTRICAL PARAMETERS, in which the voltage, voltage and current are measured in three phases of the electric motor and instantaneous values calculate the instantaneous values of power consumption. Assessment of the technical condition of the engine is carried out according to the spectrograms of losses, comparing them with spectrograms of losses obtained on a known-good engine.

The disadvantage of this method is that it allows you to evaluate the condition of only the motor itself, without taking into account the distortions that can be introduced into the operation of the electric motor by the drive mechanism and the supply converter, which limits the types of diagnosed faults.

A known method for determining the technical condition of an AC motor and a device for its implementation, described in RU 2389121 C1, H02K 15/00, G01R 31/34, METHOD FOR DETERMINING THE TECHNICAL CONDITION OF THE ELECTRIC MOTOR AND DEVICE FOR ITS IMPLEMENTATION, which consists in receiving from the sensor in the power circuit signal electric motor, its rectification and integration during the set time and measurement. After that, the integrated signal is compared with signals obtained in the same way and corresponding to the electric motor without operating hours and with the maximum operating time. The difference in the signals determines the current technical condition of the AC motor and makes an automatic forecast of the residual life.

The disadvantage of this method is that it is designed to be applied only to an alternating current electric motor, and also the influence of distortions from the drive mechanism and the supply converter on the operation of the electric motor is not taken into account. In addition, the assessment is made only on one parameter, which cannot describe the entire dynamics of the object, and the types of diagnosed faults are limited.

A known method for diagnosing mechanisms and systems with an electric drive (RU 2431152 C2, G01R 31/34, METHOD FOR DIAGNOSTIC OF MECHANISMS AND SYSTEMS WITH ELECTRIC DRIVE), taken as a prototype, in which during a given time interval the values of phase currents and electric motor voltages are recorded, after what is their decomposition into harmonic components with the measurement of the amplitude and phase of the harmonic components coming from the network. After that, the technical state is identified and the trouble-free operation resource of the diagnosed object is predicted using an artificial neural network.

The disadvantage of this method is that it does not use information about the speed and control task, which does not allow for the complete identification of the object without preliminary resource-intensive calculations of harmonic components. We have to pre-calculate the harmonic components, and the neural network classifies the malfunction according to their composition.

The technical task of this invention is the creation of a more efficient and universal method for the diagnosis of electric drives, allowing to evaluate the technical condition of the electric drive in operation at an early stage of development of defects, preventing sudden shutdowns and reducing repair costs.

The technical result consists in increasing the accuracy and reliability of diagnosis with a reduction in computational costs for assessing the state of the electric drive, through the use of a recurrent neural network as the main tool for analyzing the state of the electric drive.

The technical result is achieved by the fact that before the operation of a particular electric drive, it is preliminarily constructed its dynamic neural network model, the use of which later allows you to analyze the state of the electric drive in operation and remotely.

The essence of the invention lies in the fact that at a certain time interval, the current, voltage, speed and control task of the electric drive are measured, the parameters are converted to digital form and transferred to a personal computer for processing. A recurrent neural network, programmed and trained on a specific electric drive before its operation, reproduces the dynamics of the parameters of the electric drive, after which the result of the dynamics of the neural network model is compared with the real dynamics of the electric drive. In a faulty electric drive, a deviation of the dynamics El (t) of its parameters from the model M (t) occurs and the function of the dynamics mismatch in time Err (t) is calculated.

By the nature of the dynamics of Err (t), the technical condition is estimated and the resource of the drive is forecasted.

When considering an electric drive as a dynamic object described by an input data vector, an internal state vector, and an output data vector, we can obtain its model by using a recurrent artificial neural to identify it. Having a sufficient number of training examples, a neural network can be trained to reproduce the dynamics of an object according to the universal approximation theorem. The key condition is the presence of feedback, which allows you to remember the sequence of signals.

The structure of the neural network model is presented in figure 1.

The generalized model has the following form:

- выходной вектор, $$\overline{u}(n)$$

- входной вектор, n - дискретный момент времени, q - порядок системы.Where $$\overline{y}(n)$$

is the output vector $$\overline{u}(n)$$

is the input vector, n is a discrete time instant, q is the order of the system.

Currently, there is no exact way to determine the required number of neurons sufficient for optimal identification of an object according to the criteria of display accuracy and saving of computing resources. Experimentally, it was found the optimal use of a two-layer network with 12 neurons in the inner layer to identify the electric drive.

To collect information and carry out diagnostics, a technical complex is used (Figure 2) including: a speed sensor (DC), a current sensor (DT), a voltage sensor (DN), a measuring and computing complex (IVK), and a personal computer (PC).

At the first stage, the drive is identified. To obtain complete information, the drive starts up in all operating modes, in which it will subsequently be operated. At the same time, the current parameters with DT, the voltage with DN, the speed with DC and the control task of the electric drive, which is generated by the IVC itself, are recorded in the memory of the IVK in the operation of the current parameters with DT.

After taking the IVK data, it leads to their digital form and filters the signals of speed W, current I and voltage U to eliminate noise by the method of smoothing at several points.

where k is the number of points that are involved in smoothing the signals.

After preparing the data, the CPI transfers them to the PC to build the model. Special software builds a neural network model (Figure 1) for identification and trains the model on the set of data received from the CPI. Part of the data is backed up to test the trained model for adequacy of the dynamics display.

:After the completion of the model training stage, the diagnostic system can perform calculations in the background, evaluating the degree of coincidence of the dynamics vector of the model parameters with the dynamics of the drive parameters. In this case, the value of the dynamics discrepancy Err (t) is constantly calculated, forming a function in time over the period t, the integral estimation of the discrepancy between the dynamics IErr and the rate of change of the discrepancy is also calculated $$\frac{dErr}{dt}$$

:

Since the neural network model has some error, at the initial identification stage, the tolerances of the function ΔErr (t) and its integral estimate ΔIErr, which are taken as the normal state, are introduced.

по их совокупности за определенный промежуток времени t производится анализ состояния электропривода путем сопоставления с эталонными, заранее полученными моделями для разного типа дефектов электропривода и приводного механизма. При этом вычисляется массив коэффициентов принадлежности к каждому типу дефекта P:After calculating Err (t), IErr and $$\frac{dErr}{dt}$$

according to their totality, for a certain period of time t, the state of the electric drive is analyzed by comparing it with reference models obtained in advance for various types of defects in the electric drive and the drive mechanism. In this case, an array of coefficients of belonging to each type of defect P is calculated:

Thus, the proposed system operates in two modes: model training mode and diagnostic mode. In the training mode, it takes some time until data is taken and preparation of the formation of a trained neural network is carried out, this can take about 10 minutes due to the fact that network training is resource-intensive in terms of processor time. After training, the system operates in the diagnostic mode and calculating the current state of the dynamics and its assessment takes up little processor resources, so the method can be implemented both on a PC and entirely on the basis of the measuring and computing complex of the IVK, independent of the PC or based on the software of industrial controllers general purpose.

Claims (1)

A method for diagnosing the technical condition of an electric drive by assessing the dynamics of its parameters, including measuring and recording by a microprocessor device for a given period of time with a given frequency of motor currents and voltages, converting analog current and voltage values to digital form, transmitting digital current and voltage data to a personal computer for further processing by specialized software using an artificial neural network to identify technical the state and forecasting the resource of the trouble-free operation of the electric drive, characterized in that the microprocessor device is simultaneously measured and recorded for a predetermined period of time with a given frequency of the motor speed and the control task, the conversion of analog speed values and the control task to digital form, the transmission of digital speed and control data tasks in a personal computer for further processing with specialized software, recurren a neural network for identifying the electric drive directly by the obtained parameters, without preliminary analysis of the parameters by other methods, after identification, the neural network is a dynamic neural network model of the electric drive, while the neural network model repeats the dynamics of the electric drive in the normal operation of the electric drive, when the technical state of the electric drive changes, the dynamics mismatch neural network model and electric drive, the function of the mismatch of din is calculated Miki, the magnitude, speed and acceleration of the growth dynamics of the error functions, based on the empirical data for the different nature of faults, issued an opinion on the technical condition of the electric, the nature of the fault and the forecast of its resource.

Images