JPH0251385A - Diagnosing device for trouble of thyristorlized leonard device - Google Patents

Abstract

PURPOSE:To process the investigation of the cause of trouble automatically by previously recording the voltage/current waveforms of each section of a thyristerized Leonard device and the states of the out put of gate pulses and specifying the cause of the trouble from the data of respective section before hand recorded at the time of the generation of trouble. CONSTITUTION:A speed control amplifier SC outputs a current reference signal in response to a deviation between a speed reference signal Vr and a speed feedback signal Va. A current control amplifier CC outputs a voltage reference signal corresponding to a deviation between the current reference signal and a current feedback signal Ia. A voltage control amplifier VC outputs a control signal corresponding to a deviation between the voltage reference signal and a voltage feedback signal to a thyristor 14 through gate pulse generating circuits 32-34. The signals of each section (1)-(9) are stored in a historical memory during operation. When trouble is generated, the signals of respective section before the generation of trouble are read from the historical memory, and the position of the generation of trouble is estimated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a failure diagnosis device for a thyristor Leonard device used to drive a motor for a rolling mill.

Thyristor Leonard devices are widely used in steel mills and the like, and because most of the devices are stationary devices, they are said to be highly reliable as they do not deteriorate due to wear or the like, but of course failures do occur.

When a failure occurs in a thyristor Leonard device, particularly in its thyrisk section, it is difficult to pinpoint the location of the failure because of the large number of thyristors, the repeatability (instantaneous recovery) of the abnormality, and the like.

[Conventional technology]

Repeated abnormalities may return to normal during fault diagnosis, making it difficult to determine the location of the fault. For this,
An effective method is to constantly record the status of each part using a history memory.

For example, in Japanese Patent Application Laid-Open No. 57-162963, the gate pulses of a large number of (2 6-layer) thyrisks that supply power to a DC motor are combined into two sets, each set is made distinguishable, and the gate pulses are written into a constantly circulating memory. (When writing to the end, it returns to the beginning and writes new data on top of the old data, and repeats this process.) When an error occurs, this memory update is stopped and the state of the gate pulse for a certain period of time immediately before the stop is recorded. This will give you clues for investigating the damage site.

There are various types of failures in the thyristor Leonard device, even in the thyristor part, which can be summarized as shown in FIG. 3, for example.

The thyristor Leonard device for the rolling mill motor is, as shown in FIG.
, the field winding circuit 20, etc., and the gate pulse circuit for these sirisks includes a gate amplifier 30, a gate pulse control circuit 31, gate pulse generation circuits 32 to 34, etc. as shown in FIG. 5, and the control input is Motor 1
These include the speed reference value Vr, the actual speed value Va, the current 1d of the DC main circuit, and the voltage Vd of the DC main circuit. There are two thyristor groups, one for forward rotation and one for reverse rotation, and FIG. 5 shows one of the thyristor groups.

Since the configuration and operation of this thyristor Leonard device are known, a detailed explanation will be omitted, but the "element failure" in FIGS. 1 and 2 is a failure of each thyristor (element) of the thyristor group 14, The main ones are short circuit and open circuit (oven), the former short circuit fault is protected by fuse blowing, and the latter open circuit will result in commutation failure and overcurrent will occur, resulting in fuse blowing or DC high speed breaker CB2. Protected by actuation. The phase shifter failure shown in Figure 3 is due to a failure in the phase shifter power supply circuit shown in Figure 4, and the gate pulse output from the gate amplifier 30 becomes abnormal, which also causes commutation failure and overcurrent generation, causing the fuse to fail. Protected by fusing or by operation of a DC high-speed breaker.

The reverse phase rotation connection shown in Figure 4 is when the phase rotation direction is reversed due to a wiring error in the 3-phase AC power supply 10, power transformer 12, etc., and the power supply phase loss in Figure 5 is due to a blown phase fuse. As shown in the diagram below, these are protected by shutoff through the operation of phase loss/AC overcurrent/Sirisk overload/motor overload relay.

By the way, bilateral networks have recently been attracting attention in the computer field. The outline is explained in Figure 6.
O mark is node, l n, m, ■ is first layer, second layer,
The third and fourth layers are shown, ■1 to ■6 are inputs, 0. ~03
is the output. Each node is connected by a weighting coefficient W between the eyebrows. If we focus on one node, it will look like the same figure (b),
The human power X at this node is X=Σ Si W i.

When this human power X is applied, an output y is produced as y = f (x). Here, f is a sigmoidal function,
It has the shape shown in FIG. That is, the maximum is 1, the minimum is 0,
A value between these values is taken depending on the value of X.

As for the output, there is also one that gives a darkness value and outputs 1 if it exceeds it, and 0 if it does not. There are three layers and others. In this example, the first layer is the input layer, the fourth layer is the output layer, and the second and third layers are
The layer is an intermediate layer (hidden layer).

The number of nodes in each layer is increased or decreased as appropriate.

Neural networks are used after being trained.

That is, for input data pattern rp (=1+~16),
When you want to obtain an output data pattern 0p (=O+ ~ 0,), if you first give 1p to the input layer, the output data pattern obtained from the output layer is usually far from the desired one (the weight of each node is initially the value of the random number table), the resulting output data pattern and the desired data pattern (
For example, the deviation from O,=1.0□-0i=0) is determined, and the weighting coefficient of each node is changed so that the sum of squares of the deviation is reduced. After that, when we add the input data pattern ip to the input layer again and look at the output data pattern from the output layer, it is generally thicker than the desired one. ) As the process is repeated, the process gradually converges, and eventually, by adding the input data pattern Ip, the desired output data pattern Op (for example, 0.9.O, O, etc. in this example) can be obtained.

Maintenance and repair personnel are responsible for looking at the voltage/current waveforms at appropriate points on the equipment and deducing whether it is normal or abnormal, and if it is abnormal, where the failure is. Requires experience. Neural networks are expected to be an effective means of automating this guessing process.

[Problem to be solved by the invention]

When a failure occurs in a thyristor Leonard device, protection is provided by blowing the fuse, activating the circuit breaker, and shutting off the gate (stopping the gate pulse supply) as described above, but in order to recover, it is necessary to know the cause of the failure and remove it. However, it is not easy to know the cause of the failure, as shown in Figure 3, to trace it from the protective action to the cause of the failure, and this can only be accomplished by a skilled person who takes time and years of experience and intuition.

The object of the present invention is to automatically investigate the cause of this failure.

The present inventors continued to record the voltage/current waveforms and gate pulse output states of each part of the thyristor Leonard device at the time of failure, organized the large amount of data obtained, and determined the cause of the failure from the data of each part at the time of failure. We have successfully developed a system that can generate an output indicating the cause of a failure by inputting data from each part at the time of a failure.

(Means for Solving the Problems) Fig. 1 shows a detailed explanatory diagram of the present invention. (a) is a diagram showing the points at which signals are collected from the thyristor Leonard device in the present invention or the collected signals.■ is the output signal of the analog/log control section, which is the control signal for the gate pulse generation circuits 32 to 34.■ is the gate pulse signal that is the output of the circuit (or gate amplifier 36), and ■ is the voltage reference signal. The difference between the signal and the armature voltage of the rolling mill motor M (feedback signal of DC output voltage) becomes the input to the voltage control amplifier VC.

(2) is a current reference signal, and the difference between this and the alternating current 1a (Sirisk current feedback signal) becomes the input to the current control amplifier CC. (2) is a speed reference signal, and the difference between this and the speed feedback signal Va (2) from the speed generator TG becomes the input to the speed control amplifier SC. ■ is the AC power supply voltage.

Signals collected from each part of the thyristor Leonard device
(2) is stored in each history memory. The memories M1 to M in FIG. 1 are circular registers, but the memory format may be arbitrary. RC is the controller of the register, A/D is the sampled analog signal ■～■
This is an A/D converter that samples and digitizes the data.

Data is written into the memories M and -M in a cyclical manner, so that the memory contents are constantly updated, and when a failure occurs, writing is stopped and the memory contents at that time are retained. At the time of fault diagnosis, this retained storage content is taken into processor (microcomputer) 40 through buffer BUF and multiplexer MPX.

As shown in FIG. 1(C), the processor 40 includes a gate pulse diagnosis means 41, an AC voltage diagnosis means 42, a neural network for alternating current diagnosis 43, a neural network for DC voltage diagnosis 44, and means for holding the outputs thereof. 41a to 44a and their outputs to generate an output signal indicating the fault location. Comprehensive judgment means 50
is provided.

Most of these are software and reside in the processor's memory.

[Effect]

The signals from each part of the thyristor Leonard device are sampled, A/D converted, and then stored in the memory MI-M.
Written to q. For example, these memories M1 to M9 have 64
64 addresses (or registers)
The digital values of the samples are stored at memory addresses l, 2 .
...64, and the digital values of the 64th and subsequent samples are written back to the beginning and written to addresses 1, 2, .・・・・・・
・This is repeated thereafter. Therefore, the latest 64 samples of data are always written in the memory.

When a failure occurs, updating of memory,M,-M,stops. Therefore, 64 samples immediately before the failure occur in memory M1.
~M, will be held.

To investigate the location of the fault, the program of the processor 40 is started to fetch data from the memories M, -M, estimate the location of the fault based on the fetched data, and output a signal (message) indicating the location of the fault.

Figure 2 shows the gate pulse ■, AC voltage ■, and
Indicates AC current ■, DC voltage ■, and the waveform or state of DC current. Failure cases Nα1 and No. 1 in this figure 2
1 is related to the pulse generator, Nα2~No.

5 is related to thyrisk element, Na6 and Nα8 are related to AC power supply, Nα7 and No. 14 is related to phase shifter power supply, NO19
and no.

IO is related to the main circuit, Nα12 is related to control, and Nα1
3 is a failure other than the thyristor Leonard device. Note 1
．． Note 2. ...is as follows.

Note 1 = 1. Gate pulse is out of phase. 2° The difference in AC current is large between each pulse. 3. An open phase is seen in the DC voltage waveform. 4. The DC current fluctuates at a frequency of 50Hz.

Note 2:1. Alternating current flows rapidly at the same time as a short circuit. 2
．． A short circuit with another phase may occur after approximately 4 m5ec. 3. The DC voltage drops at the same time as a short circuit. 4. DC current is limited to OA.

Note 3:1. This waveform is similar to that of a pulse with a missing phase.

2. Alternating current is always flowing.

Note 4:1. Alternating current flows rapidly at the same time as a short circuit. 2
．． Alternating current is almost the same as an element short circuit, but the gate is cut off by the first short circuit. 3. The DC voltage drops at the same time as a short circuit, and then the AC waveform. 4. DC current is O
Narrowed down to A.

Note 5:1. The voltage waveform is similar to that of pulse open phase.

2. There is an alternating current of 1 am OA.
3. There is no phase loss in the gate pulse, but it cannot be determined from the DC voltage waveform. 4. The DC current fluctuates at a frequency of 50Hz.

Note 6:1. DC voltage waveform has 100Hz periodicity. 2
．． There is no sawtooth wave like pulse loss.

3. Alternating current is intermittent and its width is wide. (about 6m5ec). 4. DC current is hunting (100Hz cycle)
do.

Note 7:1. Same waveform as main circuit power supply phase loss. There is an open phase in the 2° pulse. 3. The alternating current has an intermittent fluctuating waveform at 100Hz.

Note 8:1. After the main circuit voltage fluctuates, it becomes a constant voltage (motor regenerative voltage). 2. The current becomes OA. 3. Since the voltage of the gate pulse decreases, the phase advances.

Note 9:1. The DC output voltage is approximately 07.

2. Short circuit current is flowing on the AC side. 3. The DC current increases in the regeneration direction after the gate is cut off.

Note 10: 1. The DC output voltage is automatically controlled to 0■. 2. Even if the output voltage is controlled to 0, a short circuit current flows in the direct current. 3. The alternating current is automatically controlled, but fluctuates at a 50Hz cycle.

Attention: 1. The DC voltage is similar to that of forward element failure. 2. The alternating current fluctuates every 50Hz.

3. The pulses are out of phase.

Note 12: 1. DC output voltage is increasing.

The Z alternating current also increases with a lag behind the increase in voltage.

3. Direct current is the same as alternating current. 46 After the gate is cut off, an AC voltage waveform is generated.

Note 13: 1. After the gate is cut off, an AC voltage waveform is generated in the DC output voltage. 2. Gate pulse has been cut. 3. When the gate is shut off, the current is cut off and no longer flows.

Note i4: 1. An alternating current voltage waveform is generated, such as an external gate cutoff. 2. The current becomes OA in half a cycle after a power outage. 3. The alternating current and direct current are synchronous.

From this Figure 2, the following can be summarized. In other words, it is necessary to check whether or not there is a pulse dropout for the gate pulse. The gate pulse diagnostic means does this and produces an output indicating the presence or absence of missing pulses. If you write 1 (or data indicating its amplitude) in memory when a gate pulse occurs, and 0 when it does not occur, when a pulse is missing, it will not be at the timing position where it should be, so this will prevent the pulse from occurring. You can see what's missing.

Regarding the AC voltage, whether it is a sine wave or a part of it is distorted, and this can be determined by performing Fourier transform to obtain harmonic components. The AC voltage diagnostic means 42 does this and produces an output indicating the presence or absence of distortion.

As for the direct current, there is not much characteristic change;
It is difficult to recognize the correlation with the failure part. Therefore, it is not used for failure diagnosis.

On the other hand, there are characteristic changes in alternating current and direct current voltage, which are correlated with the fault location. The neural networks 43 and 44 are manually inputted with the alternating current waveform and direct current voltage waveform at the time of failure, and produce outputs indicating candidates for the failure location (fault name in FIG. 2).

The number of nodes in the output layer of the neural network is 14 when outputting the fault name shown in FIG. 2. The number of nodes in the input layer is, for example, 64, and one sample is input to one note, making 64 samples in total.Therefore, assuming a sampling period of 0.4 mS, a signal waveform of a period of 25.6 mS is received. Each node in the output layer outputs a value of 1 or a value close to 1 when the fault waveform of the fault name assigned to itself is input to the input layer, and otherwise outputs a value of 0 or a value close to it.

The voltage/current waveforms in Figure 2 are during acceleration (when the steel plate is caught in the roll, under heavy load), and during startup and deceleration (
In the state where the steel plate is not caught in the roll (under heavy load), the waveform is different from that shown in FIG. 2. Therefore, two sets of neural networks 43 and 44 are prepared, one of which stores the waveform during acceleration (learns with the waveform during acceleration),
On the other hand, waveforms during startup and deceleration are stored (same as above), and when diagnosing a fault, the former is used if the fault occurs during acceleration, and the latter is used if the fault occurs during start-up or deceleration.

The waveform shown in FIG. 2 is an example, and the waveform may change slightly even if the gate pulse has the same open phase, the thyrisk element is shorted, etc. Therefore, various waveforms are used for learning so that no matter which type of waveform is input, the same output can be obtained.

The outputs of each diagnostic means and neural network are, for example, 41: no phase loss, 42: no distortion, 43: power supply phase loss: 0.
9. Phase transformer open phase is 0.8.44 is power supply open phase is 0.
9, phase transformer open phase is 0.7, etc. Based on these results taken into the latches 41a to 41d, the comprehensive judgment means 50 determines that the failure is a power supply phase failure of Nα6;
It outputs a judgment result such as ``There is a possibility that the phase transformer No. 7 has an open phase.''

Some neural networks output two values, 1 and 0, but in the example above, one outputs No, 6 outputs 1, and the other outputs Nα7 1, making it difficult to judge which is more likely. I can do it. In this regard, in the present invention, a numerical value between 0 and 1 can be output, and the degree can be grasped.

The signals ■■■■■ collected in the memory are not used for the automatic process of determining the faulty part, but the repair person can read them and use them for reference as necessary.

〔Example〕

The neural network used had a three-layer structure, with the number of nodes in the input layer being 64 and the number of nodes in the output layer being 14.

There is no method to optimally determine the number of nodes in the middle layer, but this is MX 8no' x 8 ~ 1.5L From this, 3
It was set as 3.

[Effects of the Invention] As explained above, according to the present invention, it is possible to automate the identification of a failed part of a thyristor Leonard, which greatly contributes to quick recovery, labor saving, etc.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of the principle of the present invention, Fig. 2 is an explanatory diagram showing examples of voltage/current waveforms at the time of failure, etc., Fig. 3 is an explanatory diagram of various causes of failure, and Figs. 4 and 5 are thyristor Leonard. The main circuit diagram of the device, FIG. 6 is an explanatory diagram of the neural network. Applicant Nippon Steel Corporation Patent Attorney Minoru Aoyagi Fig. 1 Fig. 2 Fig. 2 Fig. 2 Faulty manufacturing phenomenon and operation Relay protection operation 1, element failure (short circuit) Hijima-zu fused thyristor Leonard failure mode

Claims (1)

[Claims] 1. Data indicating the voltage and current waveform of the AC power supply of the thyristor Leonard device, the DC voltage waveform of the main circuit, and the timing of gate pulse generation are constantly and cyclically written, and when a failure occurs, Memory (M_1...
・M_9), a gate pulse diagnosis means (41) that inspects the gate pulse data held in the memory and generates an output indicating the presence or absence of a phase open; and a gate pulse diagnostic means (41) that inspects the AC voltage waveform data held in the memory. an alternating current voltage diagnostic means (42) that produces an output indicating the presence or absence of distortion; and an alternating current diagnostic neural network (43) that receives the data of the alternating current waveform held in the memory and produces an output indicating a candidate for a fault location. ), a neural network for DC voltage diagnosis (44) which receives the data of the DC voltage waveform held in the memory and generates an output indicating a candidate for a failure location, and these diagnostic means and neural networks (42 to 44). 1. A failure diagnosis device for a thyristor Leonard device, characterized in that it comprises a comprehensive judgment means (50) that comprehensively judges the outputs from the outputs of the devices (50) and generates an output indicating a faulty part.