JPH05140844A - Prediction system and control unit for malweaving in loom - Google Patents

Abstract

PURPOSE:To enable the optimal operation for a loom by predicting malweaving factors to be potentially caused in the future based on the current weaving situation using a neural network and by making a alarming or avoiding control based on the prediction. CONSTITUTION:Plural parameters such as weaving width, weaving texture weft kind, weft reaching time, or weft flying force each representing weaving situation like weft flying state in a loom are evaluated numerically and inputted into the microcomputer in a picking controller. Thence, at each malweaving factor determined, in advance, by a neural network, risk of developing malweaving is calculated based on the grand sum of the respective products of the inputted value and weight for the above each parameter. Thence, based on the results, alarming system and/or control unit is actuated to alter the weaving situation for the loom.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a weaving failure prediction device for a loom and a control device for the loom using the device.

[0002]

2. Description of the Related Art In a conventional loom, the weaving conditions in a warp feeding device, a weft inserting device, etc. are adjusted according to the occurrence of weaving defects such as warp breakage and weft insertion error, based on the experience and intuition of an operator. However, it requires a skilled worker,
It took a lot of work. Then, JP-A-63-751
As shown in Japanese Patent Publication No. 49, by inputting the adjustment content of the weaving condition according to the failure occurrence situation in advance to the control device, detecting the failure occurrence situation, and selecting the adjustment content corresponding thereto, There is an automatic adjustment.

[0003]

However, in this case, since adjustment is performed after the occurrence of a weaving problem, there is a possibility that a high operating rate cannot be expected and stable cloth quality cannot be obtained. Also, when adjusting the weaving conditions, if the environment such as temperature changes or the thread type changes,
There was also a fear that the adjustment could not be performed accurately.

In view of such circumstances, it is an object of the present invention to predict a weaving failure factor that may occur in the future from the current weaving state and perform warning or avoidance control.

[0005]

Therefore, according to the present invention, as shown in FIG. 1, a plurality of parameters representing the weaving condition of the loom (including the weaving condition and the weft flying condition) are digitized and input. And a means for calculating the risk of failure occurrence based on the sum of the products of the input value of each parameter and the weight for each predetermined weaving failure factor by the neural network. A weaving failure prediction device for a loom is provided.

Further, as shown in the figure, the present invention includes means for changing and controlling the weaving state of the loom based on the calculated value of the risk level of the failure using the above weaving failure prediction device. Provided is a control device for a constructed loom.

[0007]

In the weaving failure prediction device described above, the current weaving status is automatically determined by classifying the weaving status of the loom using a neural network and learning the correspondence between the classification pattern and the weaving failure factor. It is possible to predict the cause of failure that may occur in the future based on the current weaving condition while making the above determination.

[0008] Then, warning or avoidance control can be performed based on this prediction. In particular, the weaving condition of the loom can be controlled by changing and controlling the weaving condition of the loom according to the risk of each weaving failure factor as a weaving machine control device. Optimal operation becomes possible.

[0009]

Embodiments of the present invention will be described below, particularly, an example in which the occurrence of a weft insertion defect is predicted and a warning or weft insertion control is performed based on the prediction. Referring to FIG. 2, the main nozzle 1 inserts the weft Y by air jet,
Air is supplied from a pressure air supply source (not shown) through the electropneumatic proportional valve 2, the surge tank 3, and the electromagnetic opening / closing valve 4. Therefore, the injection pressure of the main nozzle 1 (the pressure of the air supplied to the main nozzle 1) can be controlled by controlling the voltage to the electropneumatic proportional valve 2, and the injection start timing and the injection end timing of the main nozzle 1 are electromagnetic on-off valves. Controlled by 4.

The weft yarn Y is drawn out from a yarn supplying body (not shown), and guided to the main nozzle 1 through a weft measuring and storing device 5. The weft measuring and storing device 5 includes a drum 7 which is rotatably supported by a tip of a hollow rotary shaft 6 which is rotationally driven by a motor (not shown) and which is held stationary, and a hollow rotary shaft 6
It has a winding arm 8 projecting diagonally from the end and communicating with it, and a solenoid-driven length measuring claw 9 that is inserted into and withdrawn from a hole provided in the peripheral surface of the drum 7, and While locking the weft Y, the weft Y is wound around the drum 7 by the rotation of the winding arm 8 and the measured length is stored. Then, the length-measuring claw 9 is retracted at a predetermined claw removal timing (weft insertion start timing), and the weft Y on the drum 7 is pulled out by the air injection of the main nozzle 1 to be weft-inserted.

The weft Y ejected from the main nozzle 1 is reed 10
Auxiliary nozzles that fly in a weft guide passage (not shown) formed by rows of recesses formed in the reed wing, and are arranged at a predetermined interval here.
It is blown in by a relay type by 11 and weft inserted. Auxiliary nozzles 11 are grouped by several (three in the figure), and from a surge tank 13 connected to a pressure air supply source via an electropneumatic proportional valve 12, air is supplied via each electromagnetic on-off valve 14. Is supplied. Therefore, the injection pressure of the auxiliary nozzle 11 (the pressure of the air supplied to the auxiliary nozzle 11) can be controlled by controlling the voltage to the electropneumatic proportional valve 12, and the injection start timing and the injection end timing of the auxiliary nozzle 11 are controlled by electromagnetic switching. Controlled by valve 14.

The weft insertion controller 15 with a built-in microcomputer receives a loom angle sensor 16 as a control input.
And the signal from the weft unwinding sensor 17 are input. The angle sensor 16 outputs a signal corresponding to the rotation angle of the loom spindle (hereinafter referred to as the loom spindle angle).

The weft yarn unwinding sensor 17 detects the passage of the weft yarn Y unwound around the drum 7 when the weft is inserted near the drum 7. Since the weft unwinding signal due to the passage of the weft Y is obtained every time one winding is unwound, assuming that one pick is n turns, it is obtained n times until the weft insertion is completed. In addition to the above, the weft insertion controller 15 includes a weft arrival sensor 18 and a weft tension sensor 19 for predicting a weft insertion failure described later.
Signals are being input from each of the above.

The weft arrival sensor 18 detects the arrival of the weft Y that has been inserted on the opposite side of the weft guide passage. The weft thread tension sensor 19 detects the tension of the weft thread Y between the drum 7 and the main nozzle 1. In addition, the weft insertion controller 15 has a man-machine interface 20
Is connected, and a human being can input the weaving width, the fabric structure, the type of weft, the thickness of the weft, the warp tension, and the like in order to predict a weft insertion defect described later.

Here, the weft insertion controller 15 is
Predetermined arithmetic processing is performed by a built-in microcomputer to determine the injection pressure P _M of the main nozzle 1 and the injection pressure P _S of the auxiliary nozzle 11, and the injection pressure of the main nozzle 1 is controlled via the electropneumatic proportional valve 2. At the same time, the injection pressure of the auxiliary nozzle 14 is controlled via the electropneumatic proportional valve 12. Further, the injection start timing T _MO and the injection end timing T _{MC of} the main nozzle 1, the claw removal timing T _CO of the measuring claw 9, the injection start timing T _SO and the injection end timing T _SC of each group of the auxiliary nozzle 11 are determined. , The opening / closing timing data is set in the solenoid controller 21.

The solenoid controller 21 controls the injection timing of the main nozzle 1 via the electromagnetic opening / closing valve 4 while monitoring the loom spindle angle based on the signal from the loom angle sensor 16, and also the solenoid-driven length measurement. The operation of the pawl 9 is controlled, and further, the injection timing of the auxiliary nozzle 11 is controlled via the electromagnetic opening / closing valve 14. That is, when the main shaft angle of the loom reaches the injection start timing T _MO , the electromagnetic opening / closing valve 4 is turned on (open) to start air injection from the main nozzle 1.

Next, when the nail removal timing (weft insertion start timing) T _CO is reached, the solenoid is turned on to extract the length-measuring nail 9, thereby starting the weft insertion. During weft insertion, the weaving machine spindle angle becomes the injection start timing T _SO for each group so that air is jetted from the auxiliary nozzles 11 of the group at that position as the tip of the weft Y runs. At this time, the electromagnetic opening / closing valve 14 is turned on (open) to start air injection from the auxiliary nozzle 11. Then, when the injection end timing T _SC is reached, the electromagnetic on-off valve 14 is turned off (closed) to end the air injection from the auxiliary nozzle 11.

Next, when the injection end timing T _MC is reached, the electromagnetic on-off valve 4 is turned off (closed) to end the air injection from the main nozzle 1. Control of the plunging timing of the measuring claw 9 is performed by monitoring the weft unwinding signal from the weft unwinding sensor 17, and when the n-th unwinding signal from the weft unwinding sensor 17 comes, the solenoid is turned on. Turn it OFF and insert the measuring claw 9 into it. As a result, the weft yarn Y is locked by the length-measuring claw 9 when the weft insertion for n turns is completed, and the weft insertion is completed.

The weft insertion controller 15 is provided with a C flag in order to give a warning based on a prediction result of a weft insertion defect described later.
A warning device 22 such as an RT display or a warning lamp for multicolor display is connected. Next, the weaving failure (weft insertion failure in this example) prediction device of the loom according to the present invention will be described.

This is constituted by a microcomputer in the weft insertion controller 15, and a plurality of parameters representing the weaving state of the loom are digitized and input, and a weft insertion defect factor determined in advance by a neural network. Each time, the risk of failure occurrence is calculated and output based on the sum of products of the input value of each parameter and the weight (coupling coefficient).

Here, there are six factors of weft insertion failure: good (no failure), jetting out, opening failure, root cutting, tip entanglement, and Kinky, and each risk is calculated. FIG. 3 shows a defect prediction neural network, which employs a hierarchical network consisting of three layers: an input layer, an intermediate layer, and an output layer.

The 15 units (I _{1 to} I ₁₅ ) of the input layer include weave width W _C , cloth design / symmetric P _S1 , cloth design / asymmetric P
_S2 , weft type / span K _Y1 , weft type / filament K
_Y2 , weft thickness S _Y , warp tension T _W , claw pulling time T _CO , warp retracting time T _O , warp entering time T _I , weft arrival time T _F1 , 1 unwinding time T _Y1 , 2 winding unwinding time T _Y2 , three-roll unwinding time T _Y3 (, ..., N-roll unwinding time T _Yn ) and weft flying tension T _WF are digitized and input.

Here, the weave width W _C is input by a person in units of millimeters, for example. Cloth / Symmetry P _S1
Is, for example, symmetric when P _S1 = 1 and asymmetric when P _S1 = 0
Further, the cloth tissue / asymmetrical P _S2 is input by a person, for example, P _S2 = 1 when asymmetrical and P _S2 = 0 when symmetrical. The weft type / span K _Y1 is, for example, K _Y1 = 1 for a span yarn, K _Y1 = 0 for other cases, and the weft type / filament K _Y2 is a filament yarn K _Y2 =
1, otherwise input K _Y2 = 0 by human. The weft thread thickness S _Y is input by a human in units of denier, for example, but a sensor of the yarn supplying section may be used.
Although the warp tension T _W is input by a human, a sensor of the backrest roller section may be used. The nail removal time T _CO is
Enter the current setting value. The warp withdrawal timing T _O and the warp entry timing T _I represent the time when the weft guide passage formed in the reed retracts from the warp opening and the time when it enters the opening. It can be known and is input by a human, but it may be detected by a sensor provided on the weft guide passage side. Weft arrival time T
_F1 is detected and input based on the signal from the weft arrival sensor 18. 1-roll unwinding time T _Y1 , 2-roll unwinding time T _Y2 , 3-roll unwinding time T _Y3 (, ..., n-roll unwinding time T _Yn ) are based on signals from the weft unwinding sensor 17. Detect and enter.
The weft flying tension T _WF is input from the weft tension sensor 19.

Therefore, I ₁ = W _C , I ₂ = P _S1 , I ₃ =
P _S2 , ..., I ₁₅ = T _WF . To the 15 units (H _{1 to} H ₁₅ ) of the intermediate layer, the signals I _{1 to} I ₁₅ are input from all the units of the input layer to each unit. In each unit of the intermediate layer, the sum of products of each input and the weight is calculated, an offset is added to this, and a response function process is performed from this value using a sigmoid function,
Output values H _{1 to} H ₁₅ are obtained in the range of ˜1.

Here, the operation example of H ₁ will be described. Each input I
The sum of the products of _i and the weight W _1i is calculated, and an offset θ ₁ is added to this to calculate Y = (ΣW _1i · I _i ) + θ ₁ (i = 1 to 15). Next, a response function process is performed using a sigmoid function to obtain H ₁ = f (Y). The sigmoid function is a monotonically non-decreasing function in the output range of 0 to 1 and reflects the nature of the saturated reaction of an actual nerve cell.

Therefore, H _j = f (( _i ΣW _ji · I _i ) +
θ _j ) (i = 1 to 15, j = 1 to 15). Output layer 6
For each unit (O _{1 to} O ₆ ),
The signals H _{1 to} H ₁₅ are input from all units in the intermediate layer. In each unit of the output layer, similarly to the above, the sum of the products of each input and the weight is calculated, the offset is added to this, and the response function processing is performed using the sigmoid function from this value, and 0 to Output values O _{1 to} O ₆ are obtained in the range of ₁ .

Here, an operation example of O ₆ will be described. Each input H
The sum of the products of _j and the weight V _6j is calculated, and an offset γ ₆ is added to this to calculate Z = (ΣV _6j · H _j ) + γ ₆ (j = 1 to 15). Next, a response function process is performed using a sigmoid function to obtain O ₆ = f (Z). Therefore, O _k
= F (( _j ΣW _kj · H _j ) + γ _k ) (j = 1 to 1)
5, k = 1-6).

Here, O _{1 to} O ₆ are numerical values of the respective degrees of danger according to the weft insertion failure factors. O ₁ is good (no defect), O ₂ is the risk of jetting out, O
₃ is the risk of poor opening, O ₄ is the risk of root breakage, O ₅ is the risk of entanglement, and O ₆ is the risk of Kinky.

As a result, on the assumption that the weights and offsets are optimized by learning described later, the value of O ₁ is the largest under the good condition, and the value of O ₂ is the most under the condition that the injection shortage occurs. It is possible to predict the occurrence of a weft insertion defect such as jetting out, for example, by increasing the size. Next, a learning method will be described.

For example, regarding each weft insertion failure factor,
In some cases, the problem may occur under certain input conditions.
Since it is known from the test, in this input condition, the output
Power O ₁~ O₆For the teacher signal T₁~ T₆Give and output
O₁~ O₆Is the teacher signal T₁~ T₆Heavy to match
This is to learn the offset and offset. That is,
When the input condition is
If you know from experience that it will not occur, enter it
Set as a teacher signal as T₁= 0, T₂= 1,
T₃= 0, T_Four= 0, T_Five= 0, T₆Give = 0 to learn
Have a lesson.

This learning procedure will be described with reference to the flowchart of FIG. This example is back propagation learning using the steepest descent method. In step 1 ( _denoted as S1 in the figure. The same applies hereinafter), weights (hereinafter referred to as coupling coefficients) V _kj , W _ji and offsets γ _k , θ _j are initialized. This may be given by a random number.

In step 2, the learning pattern I _i (I ₁
~ I ₁₅ ) is set. That is, good, jet out,
Set to a specific loom operating condition that corresponds to poor opening, broken root, tangled tip, or Kinky. Step 3
Then, the output calculation of the H _j unit is performed according to the following equation. H _j = f (( _i ΣW _ji · I _i ) + θ _j ) In step 4, the output calculation of the O _k unit is performed according to the following equation.

O _k = f (( _j ΣV _kj · H _j ) + γ _k ) In step 5, the error δ _k of the O _k unit is calculated according to the following equation. Here, T _k is a teacher signal corresponding to the set learning pattern. δ _k = (T _k −O _k ) · O _k · (1−O _k ) In step 6, the error σ _j of the H _j unit is calculated according to the following equation.

Σ _j = _k Σδ _k · V _kj · H _j · (1-H _j ) In step 7, the coupling coefficient V _kj of the H _j unit and the O _k unit is modified according to the following equation. _{V kj = V kj + α ·} δ k · H j ( where, alpha is a constant) In step 8, carried out according to the following equation to correct the offset gamma _k of O _k units.

Γ _k = γ _k + β · δ _k (where β is a constant) In step 9, the coupling coefficient W _ji of the I _i unit and the H _j unit is corrected according to the following equation. W _ji = W _ji + α · σ _j · I _i (where α is a constant) In step 10, the offset θ _j of the H _j unit is corrected according to the following equation.

Θ _j = θ _j + β · σ _j (where β is a constant) In step 11, the learning pattern I _i (I _{1 to} I ₁₅ ) is updated. In step 12, it is judged whether or not all learning patterns (that is, at least six learning patterns corresponding to good, jetting out, opening failure, root cutting, tip entanglement, and kinky respectively) are completed. If NO, Returns to step 2 and proceeds to step 13 if YES.

In step 13, the number of learning repetitions is updated (counted up). In step 14, it is determined whether or not the number of repetitions ≧ the limit number of times. If NO, the process returns to step 2, and if YES, the learning ends. This backpropagation learning method (especially the correction method of the coupling coefficient)
Will be described in more detail.

This is a neural network of three or more layers, and the square of the difference between the output of the output layer and the teacher signal
This is a learning method that modifies the coupling coefficient so as to minimize the power error. The output O _k of the output layer unit k (kth unit) is given by the following equation (1).

O _k = f ( _j ΣV _kj · H _j + γ _k ) (1) As the function f, a sigmoid function that is monotonically non-decreasing within the range of 0 to 1 is used. This function is
The feature is that the derivative can be expressed by the original sigmoid function. f (x) = 1 / [1 + exp (-2x / u ₀ )] = [1 + tanh (x / u ₀ )] / 2 (2) f '(x) = 2 · f (x) ・ [1-f (x)] / u ₀ (3) The error from the teacher signal in the output layer is δ ^k (= T _k −
O _k ), and the squared error E _{p is} to be minimized.

[0040] _{E p = Σ (T k -O} k) 2/2 ... (4) ∂E p / ∂O k = - (T k -O k) = - δ k ... (5) Therefore, the output layer unit The internal potential of k is S
Rewriting as _k (= ( _j ΣV _kj · H _j ) + γ _k ), the output becomes O _k = f (S _k ), and the influence on the output O _k with respect to the minute change of the coupling coefficient V _kj ∂O _k / ∂V _kj is
(3) by _{equation, ∂O k / ∂V kj = (} ∂O k / ∂S k) · (∂S k / ∂V kj) = f '(S k) · H j = η 1 · O k · (1-O _k ) · H _j (6) (where η ₁ is a constant).

Therefore, the influence ∂E _p / ∂V _kj of the coupling coefficient V _{kj on} the squared error E _p is calculated by the following equations (5) and (6): ∂E _p / ∂V _kj = (∂E _p / ∂O _k) · become _{(∂O k / ∂V kj) =} -η 1 · δ k · O k · (1-O k) · H j ... (7).

The updated value ΔV _kj of the coupling coefficient V _kj for reducing the squared error E _p is obtained from the following equation (8) using the steepest descent method. ΔV _kj = -α ₂ · (∂E _p / ∂V _kj ) = α ₂ · η ₁ · δ ^k · O _k · (1-O _k ) · H _j = α · δ ^k · O _k · (1- O _k ) · H _j (8) (where α is a constant and α = α ₂ · η ₁ ) In addition, in the expression (8), the error function when the pattern P is input decreases in the negative direction. , Which means to modify the coupling coefficient.

As the error δ _k of the output layer unit k, δ _k = −∂E _p / ∂S _k = − (∂E _p / ∂O _k ) · (∂O _k / ∂S _k ) = δ ^k -By using _Ok- (1- _Ok ) ... (9), Formula (8) becomes as follows.

ΔV _kj = α · δ _k · H _j (10) From the above, the error calculation formula of the O _k unit (step 5)
It ^{_{is, δ k = δ k · O}} k · (1-O k) = (T k -O k) · O k · (1-O k) ... (11) of the coupling coefficient H _j and O _k-correcting (Step 7) is as follows: V _kj = V _kj + ΔV _kj = V _kj + α · δ _k · H _j (12)

Similarly, the steepest descent method can be used for the updated value ΔW _ji of the coupling coefficient W _ji from the intermediate layer H _j to the input layer I _i . First, if the internal potential of the intermediate layer unit j is rewritten as U _j (= ( _i ΣW _ji · I _i ) + θ _j ), the output becomes H _j = f (U _j ) and the coupling coefficient W
Effect of square error E _p on small changes in _ji ∂E _p / ∂
W _ji is ∂E _p / ∂W _ji = (∂E _p / ∂S _k ) ・ (∂S _k / ∂H _j ) ・ (∂H _j / ∂ U _j ) ・ (∂U _j / ∂W _ji ) = [Σ (-δ _k) · V _{kj] · f '(U j) ·} I i = -Σδ k · V kj · H j · (1-H j) · I i ... (13) (9) Similar to the equation, if σ _j = −∂E _p / ∂U _j = _k Σδ _k · V _kj · H _j · (1-H _j ) (14) is used as the error σ _j of the intermediate layer unit j. , Equation (13) is as follows.

[0046] _{∂E p / ∂W ji = -σ j} · I i ... (15) Thus, the updated value [Delta] W _ji of the coupling coefficient W _ji, from _{(15), ΔW ji = -η 3 · ∂E} p / ∂W _ji = η ₃ · σ _j · I _i (16) (where η ₃ is a constant) can be written.

From the above, the error calculation formula (step 6) of the H _j unit is as follows: σ _j = _k Σδ _k · V _kj · H _j · (1-H _j ) ... (17) Coupling coefficient of I _i and H _j The correction formula (step 9) is as follows: W _ji = W _ji + ΔW _ji = W _ji + α · σ _j · I _i (18) (where α is a constant and α = η ₃ ).

In the above equations, the error E _p is minimized for one input / output pair, and E _t = _p Σ for various input / output combinations (indicated by P). _{k Σ (T pk -O pk)} 2/2 = p ΣE p ... (19) must determine the error function, gradually reduce the error E _p in the training of each pattern P, as a whole error function E _t
To minimize.

As the learning mode, as described above, in addition to the mode of learning by setting the loom operating state corresponding to the cause of failure, when learning, the loom is appropriately operated, and when a failure occurs, It is also possible to adopt a mode in which a human inputs a defect factor (discharging, etc.) using an operation button or the like to give a teacher signal to learn the correspondence between the driving state at that time and the defect factor.

Next, a warning device and a loom control device using the above-described failure prediction device will be described. FIG. 5 is a flowchart showing a processing procedure when the device is used as a warning device. In step 21, the maximum output is detected from the outputs O _k (O _{1 to} O ₆ ) of the neural network, and this is set as O _MAX .

In step 22, it is judged whether or not O _MAX = O ₁ .
Return to 21. If no, go to step 23. Step
At 23, it is determined whether or not O _MAX = O _{2, and} if YES, it is determined that the risk of injection shortage is high and there is a tendency for injection shortage, and the process proceeds to step 24, and the warning device 22 displays a screen to that effect. Turn on the warning light. If no, go to step 25.

In step 25, it is determined whether or not O _MAX = O _{3, and} if YES, it is determined that there is a high risk of opening defects and there is a tendency for opening defects, the process proceeds to step 26, and a screen display / warning lamp to that effect is displayed. Lights up. If no, go to step 27. In step 27, it is judged whether or not O _MAX = O ₄ ,
In the case of YES, it is determined that the risk of root breakage is high and there is a tendency for root breakage, and the process proceeds to step 28, and the screen display / warning light to that effect is turned on. If no, go to step 29.

In step 29, it is judged whether or not O _MAX = O _{5, and} if YES, it is determined that the risk of entanglement is high and that the entanglement tends to occur. Then, the process proceeds to step 30, and the screen display / warning light to that effect is displayed. Lights up. If no, go to step 31. If you proceed to step 31, inevitably O _MAX = O
_6, and is high Kinky risk, This includes a due to root out tendency, because there is a due to previously entangled trend, here, either closer or, roots out risk O ₄ And the degree of entanglement risk O ₅ are compared to make a determination.
If the risk O ₄ of root breakage is higher, the process proceeds to step 32, and the screen display / warning light to that effect is lit. If the tip entanglement risk O ₅ is higher, step
Go to 33 and turn on the screen display / warning light to that effect.

In this case, the operator knows the cause of failure predicted by the screen display or the warning light, and appropriately changes the weaving condition based on this. FIG. 6 is a flow chart showing a processing procedure when it is used as a control device for a loom. In step 41, the maximum one is detected from the outputs O _{1 to} O ₆ of the neural network,
_Let this be O _MAX .

In step 42, it is determined whether or not O _MAX = O ₁ , and if YES, it is determined that it is good,
Return to 41. If no, go to step 43. Step
At 43, it is determined whether or not O _MAX = O _{2, and} if YES, it is determined that the risk of injection shortage is high and there is a tendency for injection shortage, and the process proceeds to step 44, in order to shorten the preceding injection period, the main nozzle 1 The injection start timing T _MO of the auxiliary nozzle 11 is delayed by 2 ° with respect to the current set value at the main shaft angle of the loom, and the injection start timing T _SO of the auxiliary nozzle 11 is also delayed by 2 °. If no, go to step 45.

In step 45, it is determined whether or not O _MAX = O _3. If YES, it is determined that the risk of defective opening is high and there is a tendency for defective opening, and the process proceeds to step 46 to prevent the weft from being caught in the warp. In order to do so, the claw removal timing (weft insertion start timing) T _CO of the measuring claw 9 is delayed by 2 °. If no, go to step 47. In step 47, it is determined whether or not O _MAX = O _{4, and} if YES, the root cutting risk is high and the root cutting tends to occur.
Injection pressure P _S of 2% less than the current set value (× 0.9
8) and the injection end timing T _MC of the main nozzle 1 is set to 2
° Advance. If no, go to step 49.

In step 49, it is determined whether or not O _MAX = O _{5, and} if YES, the risk of tip entanglement is high, and the tip entanglement tends to occur, the process proceeds to step 50, and the injection pressure P _{M of the} main nozzle 1 is determined. Is increased by 5% (× 1.05), and the injection pressure P _S of the auxiliary nozzle 11 is increased by 5% (× 1.05). N
If O, go to step 51. When the process proceeds to step 51, O _MAX = O ₆ is inevitable, and the Kinky risk is high. However, there are two types, one is due to the root cutting tendency and the other is due to the tip entanglement tendency. , Here, which one is closer to the other, the risk of root cut O ₄ and the risk of entanglement O
Determined by comparing with ₅ .

When the risk of root breakage O ₄ is higher,
In step 52, the injection end timing T _MC of the main nozzle 1 is advanced by 2 ° and the injection end timing T _SC of the auxiliary nozzle 11 is delayed by 2 °. If the tip entanglement risk O ₅ is higher, the process proceeds to step 53 and the injection pressure P of the main nozzle 1 is increased.
_M is increased by 5% (× 1.05) and the injection pressure P _S of the auxiliary nozzle 11 is increased by 5% (× 1.05).

Therefore, predicted failure factors can be automatically dealt with, and optimum operation of the loom can be achieved. In addition to the logic, the control target and the control amount thereof may be calculated by fuzzy reasoning.

[0060]

As described above, according to the present invention, a neural network can be used to predict a weaving failure factor that may occur in the future from the present weaving state. Therefore, warning or avoidance control can be performed based on this prediction. In particular, optimal control of the loom can be performed by changing and controlling the weaving state of the loom according to the degree of danger of each weaving failure factor.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing the configuration of the present invention.

FIG. 2 is a system diagram of a weft insertion control device showing an embodiment of the present invention.

FIG. 3 is a diagram showing a defect prediction neural network.

FIG. 4 is a flowchart showing a learning procedure.

FIG. 5 is a flowchart when used as a warning device.

FIG. 6 is a flow chart when used as a control device for a loom.

[Explanation of symbols]

 1 Main nozzle 2 Electro-pneumatic proportional valve 4 Electromagnetic on-off valve 9 Measuring claw 11 Auxiliary nozzle 12 Electro-pneumatic proportional valve 14 Electromagnetic on-off valve 15 Weft insertion controller 16 Loom angle sensor 17 Weft unwinding sensor 18 Weft arrival sensor 19 Weft tension sensor 20 Man-machine interface 21 Solenoid controller 22 Warning device

Claims (2)

[Claims] 1. A means for numerically inputting a plurality of parameters representing a weaving state of a loom and a neural network for each predetermined weaving failure factor to obtain a sum of products of an input value of each parameter and a weight. A weaving failure prediction device for a loom, which is configured to include a means for calculating a risk of failure occurrence based on the above. 2. A means for numerically inputting a plurality of parameters representing a weaving state of a loom and a neural network for each predetermined weaving failure factor to obtain a sum of products of input values of the respective parameters and weights. A loom controller configured to include a means for calculating the risk of failure occurrence based on the above, and a means for changing and controlling the weaving state of the loom based on the calculated value.