JPH07164199A - Quality controlling method of press machine and its device - Google Patents

Abstract

PURPOSE:To improve the quality and reliability of a work in a work quality controlling method and its device in a press machine. CONSTITUTION:Pressing signals of a pressure sensor 3 in a press machine are inputted into a computer 5 having a neural net-work device 6, a recording device 7 of the computer displays that the work is normal or defective and together a controller connected to the computer operates and controls the press machine. Loading wave shape pattern of a normal products wave shape group of the work or a defective products wave shape group and the operational characteristics of a neural circuit net of wave shape understanding and the operation controlling information of the press machine 1 are inputted into the neural net-work device 6 while learning, the computer 5 calculates the loading wave shape pattern matching to the operational characteristics following passing of time during operating based on learning of the neural net-work device 6, the controller a operates ana controls the press machine 1, the recording device 7 displays judging the normal/defective quality, and the operation is stopped at the time displaying that it is defective.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a work (workpiece) in a press working machine for press working such as forming, shearing, bending, and compression working, to a reference work learned in advance and a change with time of the press working machine. The present invention relates to a quality control method and apparatus for a press machine that performs press processing in consideration of accompanying operating characteristics.

[0002]

2. Description of the Related Art A press-working machine for press-working a work of this type which has been already proposed is constructed as shown in FIG.

That is, in FIG. 9, an anvil b is provided in the lower part of a press main body a in a press working machine, and a lower die c is placed on the anvil b.
Further, a displacement sensor (also referred to as a bottom dead center sensor) d is laid on the lower die c, and a press bottom dead center management device (also referred to as a comparison determination circuit) e is connected to the displacement sensor d. A printer (recording device) f and a personal computer (computer) g each having a display function are connected to the press bottom dead center management device e. Further, as shown in FIG. 10, the comparison / determination circuit e amplifies the displacement signal input from the pressure sensor d by an amplifier k via a trigger j, and thereafter sequentially compares the displacement signals with the comparison circuit m-A / D.
Displacement is displayed on the load display o through the converter n, and this is input to the personal computer g and also input to the comparators p1 and p2, respectively, and the comparators p1 and p2 are set to the bottom dead center set value (pressure). The upper limit setting value and the lower limit setting value of
A work w that exceeds the range between the upper limit setting value and the lower limit setting value is detected as a defect in press working.

On the other hand, as a management device using a load sensor, a ram member h is fitted on the upper part of the press body a so as to be able to move up and down, and an upper die i is mounted on the lower part of the ram member h. It is provided so as to press the work w in cooperation with the lower die c.

Therefore, in the above-described press machine, when the pressure sensor d detects the load signal when the predetermined work w is pressed, the peak load signal of the pressure sensor d is transmitted to the press load management device e. As shown in the graph of FIG. 11, it is controlled that the press working is performed within a range of a fixed upper limit set value and a fixed lower limit set value. Moreover, FIG.
As shown in the time chart of FIG. 2, the press load management device e observes the press peak load and controls the quality of the press processing machine that performs press processing.

Alternatively, the pressure sensor d captures a temporal change in the press load and triggers a pattern judgment circuit (not shown), whereby a normal press load reference signal recorded in the pattern judgment circuit. The waveform pattern and the waveform pattern of the detection signal of the crimping force of the pressure sensor d are compared, and the work amount as an integrated value and the statistical value of the deviation from the vertical allowable value of the good product waveform are determined as the difference, and the press load is applied. There is also a device that sends an alarm by issuing an abnormal signal when the waveform pattern of is abnormal.

[0007]

However, the above-described press working machine forms the work w into a predetermined shape while detecting the press load when the work w is pressed by the pressure sensor d. Due to such pressing, it is not only difficult to effectively and properly detect each defective work w, but it is also difficult to control the quality of the press machine itself.

Further, generally, the load detection has more sensitive accuracy than the displacement detection, and there is a recognition technique by the differential method.

As described above, the above-described press working machine simply detects the displacement of the work w at the bottom dead center of the work w or the press load with the pressure sensor d, and visually recognizes the defective product. There is a problem in reliability in setting the allowable range which is the standard for the quality judgment.

On the other hand, in the above-mentioned press working machine, since the upper die i is connected to the ram member h to form the C-shape, the bottom dead center of the upper die i gradually increases with the lapse of time from the start of operation. The temperature of the press machine itself tends to change and the jaws open depending on the temperature and the number of impacts.

That is, as shown in the graph of FIG. 13, in the above-described press machine, the relationship between the stress and the stress in the cyclic stress-strain diagram of cast iron (frame material) depending on the number of press processes after the start of operation. As is clear from
Observed as variations with hysteresis. In addition, since the above-described press working machine is formed into a C-shape, the jaws of the press body a are opened very slightly with the lapse of time from the start of operation. Further, as shown in the graph of FIG. 14, the press machine described above has a great influence from the material of the press body a itself, as is clear from the comparison diagram of the vibration damping ability of cast iron, spheroidal graphite cast iron and steel. Is receiving.

In order to solve the above-mentioned problems, the present invention calculates by a computer based on a press pressure signal of a pressure sensor in a press working machine and learning of a neural network, inputs this to a control device, and this control device PROBLEM TO BE SOLVED: To provide a quality control method of a press machine and an apparatus thereof for controlling the operation of the press machine by using, to promptly process and eliminate the defective work due to the press work, and to improve the quality and reliability of the work. With the goal.

[0013]

According to the present invention, a crimping signal of a pressure sensor in a press machine is input to a computer equipped with a neural network device, and a display device of this computer displays a work quality judgment result. In a work quality control device for controlling operation of the press machine by a control device connected to a computer, a press load waveform pattern and waveform recognition by a non-defective product waveform group or a defective product waveform group of the work is previously provided to the neural network device. Input while learning the operation characteristics based on the operation management information of the neural network and the press processing machine, and load waveforms by the non-defective product waveform group and the defective product waveform group according to the above operation characteristics with the passage of time during operation. The pattern is played on a computer based on the learning of the above neural network device. And then controls the operation of the above press machine with the control device, the quality determination of the quality displayed on the recording apparatus, a quality control method of a press machine to shut down when the defective display.

According to another aspect of the present invention, a press machine is provided with a trigger sensor and a pressure sensor, a charge amplifier unit is connected to the pressure sensor, and a computer having a CPU and an internal memory for the charge amplifier unit and the trigger sensor is provided. Connected to this computer, a neural network device for learning and recognizing the load waveform pattern of the pressing process is connected, and the I / O of the computer is connected.
Is connected to a recording and display device for recording and displaying the load waveform pattern and the quality judgment, and the controller for controlling the operation of the press machine is connected to the A / D converter board unit of the computer.

[0015]

In accordance with the present invention, the neural network device is preliminarily trained to learn the load waveform pattern based on the non-defective product waveform group or the defective product waveform group of the workpiece and the neural network for waveform recognition, and the operation characteristics based on the operation management information of the press machine. While inputting and recording in the recording device, the load waveform pattern by the good product waveform group of the work and the defective product waveform group according to the operation characteristics with the passage of time during operation is calculated by the computer based on the learning of the neural network device. This is a quality control method of a press machine, in which the press machine is operated and controlled by the control device, the quality judgment and the load waveform pattern are displayed on the display device, and the operation is stopped when a defect is displayed.

[0016]

The present invention will now be described with reference to the illustrated embodiments.

In FIG. 1, reference numeral 1 is a press machine for press-working a work w.
For example, a trigger sensor 2 including a photo sensor and a pressure sensor (also referred to as a load sensor) 3 including, for example, a load cell are attached to the inside of the trigger sensor 2. The trigger signal 2a of 2 is input to the computer 5 described later. A charge amplifier unit 4 for amplifying an output voltage is connected to the pressure sensor 3, and a computer 5 is connected to the charge amplifier unit 4 and the trigger sensor 2. Further, in the computer 5, a CPU (central processing unit) 5a and various reference works (reference signal waveform patterns of normal press pressure) having different shapes and sizes of various works are stored and stored. The memory 5b, the interface 5c, the communication unit 5d, and the I / O 5e are incorporated. Furthermore, a neural network device 6 is connected to the communication section 5d, and when the neural network device (recognition storage device) 6 presents a work example by waveform analysis by the computer 5, a desired waveform analysis conversion is performed. I am learning and memorizing it.

In particular, the neural network device 6 described above.
Is configured in advance so that the neural network of the load waveform pattern and the waveform recognition by the non-defective crimping waveform group of the work w or the defective crimping waveform group can be learned. The operating characteristics of the processing machine 1 are learned while inputting, and during operation, the CPU 5a in the computer 5 and various workpieces having different shapes and sizes are pressed as standard workpieces (normal workpieces). In the internal memory 5b storing the pressure reference signal waveform pattern), the interface 5c, the communication unit 5d, and the I / O 5e, the good product crimping waveform group of the work and the defective product crimping waveform group are described above according to the operating characteristics with the aging. The press machine 1 is operated and controlled by a control device 8 which will be calculated based on learning of the neural network 6 and will be described later.

That is, as shown in FIG. 6, the neural network 6 device is similar to the curve A according to the reference signal waveform pattern of normal pressure so that the pressure sensor 3 indicates the appropriate range of the pressure of the work w. The neural network device 6 detects the pressure with a degree of allowable degree, and the neural network device 6 performs the load waveform pattern by the non-defective product press waveform group of the work w or the defective product press waveform group and the neural network of the waveform recognition and the press processing. It is composed of an input layer for learning and inputting operation characteristics of the machine 1 with time, an intermediate layer for storing vectors, and an output layer for summing the vectors.

On the other hand, a recording device 7 is connected to the I / O 5e of the computer 5, and the recording device 7 is composed of an external recording device 7a and a display device 7b. As shown in FIGS. 2 to 5, recording and reproducing of the quality of the work are displayed.

That is, the display device 7b is shown in FIGS.
As shown in, the characteristic curve of the time (t) of the operating characteristics (dynamic characteristics) of the press machine 1 and the number of times of press working is displayed.

That is, as shown in FIGS. 2 and 3,
The press machine 1 presses the work w at the first time, the curve a becomes the highest curve, at the fifth time, the curve b, at the tenth time, the curve c, and at the thirtieth time. Is the lowest curve d. Each of these curves a,
After the start of operation, b, c, and d gradually become a downward curve, the press machine 1 is damping stabilized, and the neural network device 6 learns the operation characteristics of the press machine 1 with the lapse of time. Are inputting. Further, in the press machine 1, as an example, as shown in FIG. 4A in the case of crimping, as shown in FIG. 4 (A), after the start of operation, the first initial input crimping type waveform a ′ 1 is identified for the first time. The separated pressure bonding type waveform a1 is learned and displayed, and the above-mentioned press processing machine 1
4B, as shown in FIG. 4B, displays the first discrimination and separation crimping waveform a1 and the damping-stabilized thirtyth discrimination separation and crimping waveform d1. 6 also learns the operating characteristics of the press machine 1 with the lapse of time.

Further, the curve a1 in FIG. 5 is the discriminative separation crimping type waveform of the first non-defective product, and the curve d1 in FIG. 5 is the learning stabilized display of the 30th discrimination separating crimping type waveform. The curve a3 in FIG. 5 is obtained by learning the first resin biting failure waveform. Further, the curve e in FIG. 5 is a learned and displayed defective waveform synthesis curve.
Here, the defective waveform synthesis curve e is e = (d1 / a1) * a3.

The neural network 6 also learns the operating characteristics of the press machine 1 with the lapse of time from the curves shown in FIG.

In particular, as shown in FIG. 6, the neural network device 6 includes the neural network for the load waveform pattern and the waveform recognition by the non-defective press waveform group of the work or the defective press waveform group of the work and the press processing machine 1. In addition to learning the driving characteristics with the passage of time, the additional learning is performed to improve the accuracy, and as shown by each curve in FIG. 5, the fine adjustment of the discrimination function can be performed by the waveform input and the synthetic learning.

Here, the synthetic learning is, as an example, as shown in the graph of FIG. 3, by recording the attenuation waveform of the non-defective press working step, as shown by the curves a3 and e of FIG. By synthesizing defective crimp waveforms, the attenuation rate at each detection point of non-defective product waveform sampling is calculated, and by multiplying by the first crimp waveform of defective product waveform sampling, defective product waveform sampling over time can be performed. Pseudo-synthesis is performed, and by learning this with the neural recognition software of the neural network device 6 in the same manner as the defective product crimping waveform group of the work w, the accuracy of defective product detection during aging and after stabilization is increased. It is also possible.

On the other hand, as shown in FIGS. 3 to 5,
The recorded waveform and the recognition state can be verified by the recording device 7 and displayed in an overlapping manner to verify the discrimination and separability, and at the same time, the waveform sample can be displayed in the superimposed state on the screen of the recording device 7. The ring and recorded waveform can be modified.

On the other hand, the controller 8 is connected to the A / D converter board unit 5f of the computer 5 so as to control the operation of the press machine 1.

That is, when the control device 8 detects and displays the defective work w, the operation of the press machine 1 is instantaneously stopped.

The operation of the present invention will be described below. Therefore, the pressure bonding signal of the pressure sensor 3 in the press machine 1 is input to the computer 5 equipped with the neural network device 6, the quality of the work w is displayed on the display device 7b of the computer 5, and the computer 5 is connected to the computer 5. In the work quality control device for controlling the operation of the press machine 1 by the control device 8, the neural network device 6 previously detects the load waveform pattern and the waveform recognition by the good product crimp waveform group of the work w or the bad product crimp waveform group. The operating characteristics of the press machine 1 with time change are learned and input with the neural network.

In this case, as shown in FIGS. 3 to 4 (b), the neural network device 6 is accompanied by a good product crimping waveform group of a work, a defective product crimping waveform group, and the lapse of time of the press machine 1. Learning driving characteristics.

This will be described with reference to the flow chart shown in FIG. A reference work (reference signal waveform pattern of normal pressure) for pressing various works w having different shapes and sizes of various works w is stored in advance, and operating characteristics of the press machine 1 with time change. The operation of inputting while learning corresponds to steps s1 to s8 of the flowchart shown in FIG.

Next, in steps s9 to s15 of the flow chart shown in FIG. 7, the neural network device 6 performs the above-mentioned neural network for the recognition of the load waveform pattern and the waveform pattern by the non-defective press waveform group of the work or the defective press waveform group of the work. And inputting while learning the operating characteristics of the press machine 1 with time.

Further, when the press machine 1 is in operation, the computer 5 receives a good product press waveform group and a defective product press waveform group of the work in accordance with the operation characteristics with the aging, based on the learning of the neural network device 6. And the operation of the press machine 1 is controlled by the control device 8.
The quality judgment is displayed on the display device 7b of the recording device 7, and the quality control of the press machine 1 which is stopped when the defect is displayed is controlled.

This will be described with reference to the flow chart shown in FIG. 7, which corresponds to steps s16 to s21.

Therefore, according to the present invention, when the work w on the anvil 1a of the press machine 1 is pressed, the trigger sensor 2 starts sampling the output waveform signal that determines the display content of the time-based waveform. That is, the trigger signal 2a of the trigger sensor 2 is input to the computer 5 and the pressure sensor 3 detects the pressing force while the timing of starting the press, and the pressure signal of the pressure sensor 3 is transmitted to the charge amplifier unit 4. Input to and amplify the output voltage.

That is, the press process output waveform signal of the pressure sensor 3 is converted into a voltage in the charge amplifier unit 4.
Here, the charge amplifier unit 4 has a function of managing the pressing process, and is set so that the normal output peak value of the pressing process is about 70% of the predetermined maximum pressing force.

Further, the pressure signal of the pressure sensor 3 which has been voltage-converted in the charge amplifier unit 4 is input to the computer 5, and thereafter C of the computer 5 is converted.
The PU 5a performs a comparison calculation with a reference work example by waveform analysis by the internal memory 5b, and inputs this to the control device 8 via the TTL signal 8a of the A / D converter board unit 5f to operate the press machine 1. Control.

That is, the control device 8 has a defective work w.
When is detected and displayed, the operation of the press machine 1 is instantaneously stopped. That is, when the quality is out of the allowable setting range of the goodness-of-goods similarity, which is a reference for quality judgment, the control device 8 causes the operation of the press machine 1 to stop urgently. Further, the control device 8 finely adjusts the bottom dead center of the ram.

On the other hand, the computer 5 has its communication section 5
Input to the neural network device 6 through d.
Then, the neural network device 6 presents each work example by the waveform analysis and imitates and learns the conversion of the desired waveform analysis. Furthermore, the computer 5 displays recording and reproducing of the quality of the work from the I / O 5b to the recording device 7.

Next, in another embodiment of the present invention shown in FIG. 8, a tachometer 9 such as a rotary encoder is attached to the press machine 1 and the tachometer 9 is connected to the computer 5. By doing so, the displacement-like change in the pressure signal is captured as a pattern.

[0042]

As described above, according to the present invention, the neural network device is provided with a neural network for recognizing a load waveform pattern and a waveform by a non-defective product waveform group of a work or a defective product waveform group and the operation of the press machine in advance. Input while learning the operating characteristics based on the management information, and based on the learning of the neural network device, the load waveform patterns by the good product waveform group and the defective product waveform group of the work according to the operating characteristics with the passage of time during operation. Since the computer controls the operation of the press machine with the control device, the quality judgment of the quality is displayed on the recording device, and the operation is stopped when the defect is displayed. Not only can it be eliminated, but the quality of the work, the verification of the load waveform pattern, and the reliability can be improved.

Further, in another invention, a trigger sensor and a pressure sensor are attached to a press machine, a charge amplifier unit is connected to the pressure sensor, and the charge amplifier unit and the trigger sensor are provided with a CPU and an internal memory. A computer is connected, a neural network device for learning and recognizing the load waveform pattern of the press working process is connected to the computer, and a recording device for displaying the load waveform pattern and the quality judgment is connected to the I / O of the computer. Since a control device is connected to the A / D converter board unit of the computer to control the operation of the press machine, fine adjustment and additional learning can be performed with respect to changes over time, recording, learning omissions, etc. Not only can you learn the defective press waveform group of the work, By reducing the less defective an excellent effect such as it can be improved production efficiency.

[Brief description of drawings]

FIG. 1 is a block diagram of a quality control method for a press machine and an apparatus thereof according to the present invention.

FIG. 2 is a front view of a display unit of a display device used in the present invention.

FIG. 3 is a graph showing operating characteristics of the press machine of the display unit of the display device.

FIG. 4 is a graph showing the relationship between time and voltage of the press machine of the display unit of the display device.

FIG. 5 is a graph showing the relationship between time and voltage of the press machine of the display unit of the display device.

FIG. 6 is a diagram for explaining learning of a neural network device used in the present invention.

FIG. 7 is a flowchart of a work quality control method of the present invention.

FIG. 8 is a diagram showing another embodiment of the present invention.

FIG. 9 is a block diagram of a conventional quality control device for a press machine.

FIG. 10 is a block diagram of a press load management device incorporated in a conventional quality control device.

FIG. 11 is a graph showing the relationship between the press pressure and time of the quality control device of the conventional press machine.

FIG. 12 is a time chart of a quality control device for a conventional press machine.

FIG. 13 is a graph showing the relationship between strain and stress in a conventional press machine.

FIG. 14 is a graph showing a comparison of vibration damping capabilities of materials of a conventional press machine.

[Explanation of symbols]

 1 Pressing Machine 2 Trigger Sensor 3 Pressure Sensor 4 Charge Amplifier Unit 5 Computer 6 Neural Network Device 7 Recording Device 8 Control Device

Claims (2)

[Claims] 1. A press-bonding signal from a pressure sensor in a press machine is input to a computer equipped with a neural network device, and the quality of the work is displayed on a recording device of the computer, and the press is controlled by a control device connected to the computer. In a work quality control device that controls the operation of a processing machine, a neural network for load waveform pattern and waveform recognition by a non-defective product waveform group or a defective product waveform group and the operation management of the press processing machine in advance in the neural network device. Input while operating the operating characteristics by learning, and based on the learning of the neural network device, the load waveform pattern by the non-defective product waveform group or the defective product waveform group of the work according to the operating characteristics with the passage of time during operation. The computer controls the press machine using the controller. Controlled, the quality determination of the quality displayed on the recording apparatus, the quality control method of the press machine, characterized in that the shutdown during display failure. 2. A press sensor equipped with a trigger sensor and a pressure sensor, a charge amplifier unit connected to the pressure sensor, a CPU connected to the charge amplifier unit and the trigger sensor, and an internal memory. A computer, a neural network device connected to the computer for learning and recognizing a load waveform pattern in a press working process,
A recording device and a display device connected to O for recording and displaying a load waveform pattern and quality judgment, and a control device connected to the A / D converter board unit of the computer so as to control the operation of the press machine. A quality control device for a press machine, characterized by comprising: