JP7452055B2 - Winch drum irregular winding detection device - Google Patents

Description

TECHNICAL FIELD The present invention relates to a winch drum irregular winding detection device for detecting a state in which a wire rope wound around a winch drum is irregularly wound.

The wire rope used to hang the crane's load is subject to a large amount of tension, so if the crane continues to operate with the wire rope wound around the drum in an unaligned state, the wire rope will be damaged and extremely dangerous. Therefore, a method of detecting the irregular winding state has been proposed.

Patent Document 1 describes a display control device for a surveillance camera of a mobile crane. In the display control device of Patent Document 1, a plurality of surveillance cameras are provided to monitor blind spots such as behind the upper revolving structure and on the opposite side of the driver's cab, and to monitor the state of irregular winding of the winch drum. A monitor is installed in the driver's cab and images from each surveillance camera are switched and displayed.

The rope irregular winding detection device of Patent Document 2 is a rope winding device that winds a rope into a rope groove machined in a winding drum, and includes a screw feeding device whose feeding direction is parallel to the winding drum; A power transmission device is provided between the screw feeding device and has a reduction ratio that gives the sent object the same speed as the horizontal speed of the rope when winding and unwinding the rope. The rope detector is moved at the same speed as the horizontal movement speed of the rope that is caught in the rope groove and moved horizontally, and the rope is moved horizontally to prevent the rope from coming off the rope groove. When a change occurs in the horizontal speed of the rope, it is detected that the rope has deviated from the center of the rope detector and the rope detector has started to wind irregularly.

In the irregular winding detection device of Patent Document 3, a detection bar faces the outer peripheral surface of a hoisting drum on which a rope is symmetrically wound at both ends in the axial direction, and the detection bar detects winding as the rope is wound in stages. A winding irregular winding detection device for a hoisting drum that detects irregular winding by moving away from the outer peripheral surface of the upper drum, which includes a central portion of the hoisting drum where no rope is wound, and a detection bar facing the central portion of the hoisting drum. The detection bar is characterized in that the center portion of the detection bar has a larger diameter than both end portions sandwiching the center portion so that the distance from the center portion is smaller than the diameter of the rope.

Japanese Patent Application Publication No. 2004-137035 Japanese Patent Application Publication No. 58-6892 Japanese Patent Application Publication No. 04-129997

In the method disclosed in Patent Document 1 in which images from a surveillance camera are displayed on a monitor in a driver's cab, an operator visually determines the winding state of a wire rope. Since the operator does not always see the image of the winch drum, there is a possibility that the occurrence of a winding condition may be overlooked.

The rope irregular winding detection device of Patent Document 2 and the rope irregular winding detection device of Patent Document 3 detect irregular winding by operating a switch using a mechanical action. However, both devices assume that only one layer of wire rope is wound on the drum. The irregular winding detection device of the patent document cannot be applied to a crane with a large lifting height in which wire rope is wound in multiple layers on a drum.

The present invention was made in view of the above-mentioned situation, and it is an object of the present invention to detect the occurrence of random winding, regardless of the number of layers of wire rope wound around a drum.

The winch drum irregular winding detection device according to the first aspect of the present invention includes:
an imaging device that generates image data by photographing a wire rope wrapped around a drum;
The wire rope includes a trained model of a neural network that analyzes image data generated by the imaging device, and the wire rope is in a normal winding state where the wire rope is wound in alignment on the drum, or the wire rope is wound around the drum. a random winding determination unit that determines whether the winding is in a random winding state where the windings are not aligned;
Equipped with
The random winding determination unit is
a compression neural network that obtains a latent variable of the state in which the wire rope is wound around the drum from the image data;
a restoration neural network that generates reproduced data of the same dimension as the image data from the latent variables acquired by the compression neural network;
a comparison unit that compares the image data input to the compression neural network and the reproduction data generated by the restoration neural network to detect different parts of the image;
including;
The compression neural network and the restoration neural network use, as learning data, image data generated by photographing the wire rope in a normal winding state where the wire rope is wound in alignment on the drum, and combine the image data and the reproduction data. It is a trained model of a neural network that is machine-trained to keep the difference within a specified range.
The irregular winding determination unit determines that the image data is in the irregular winding state when a difference between the image data and the reproduced data generated by the learned model exceeds a predetermined range.

The winch drum irregular winding detection device according to the second aspect of the present invention includes:
an imaging device that generates image data by photographing a wire rope wrapped around a drum;
The wire rope includes a trained model of a neural network that analyzes image data generated by the imaging device, and the wire rope is in a normal winding state where the wire rope is wound in alignment on the drum, or the wire rope is wound around the drum. a random winding determination unit that determines whether the winding is in a random winding state where the windings are not aligned;
Equipped with
The random winding determination unit is
A neural network uses image data of a normal winding state in which the wire rope is wound in alignment on the drum, and image data of a disorderly winding state in which the wire rope wound around the drum is not aligned, as learning data. a trained model of a neural network that is subjected to machine learning to classify the image data into the normal winding state and the irregular winding state;
The trained model of the neural network classifies the image data generated by the imaging device into each state of irregular winding including rope gap, run-up, multiple run-up, and rope slack, and the normal winding state. Outputs the probability that
The random winding determining unit determines whether the image data is in the normal winding state or the irregular winding state based on the probabilities of being classified into the respective classifications output by the trained model .

According to the present invention, it includes a trained model of a neural network that analyzes image data taken of a wire rope wound around a drum, and determines whether the wire rope is normally wound or irregularly wound. The occurrence of random winding can be detected regardless of the number of layers of wire rope wound around the drum.

A block diagram showing the configuration of a winch drum irregular winding detection device according to an embodiment of the present invention Block diagram showing the functional configuration of the image analysis unit according to Embodiment 1 A diagram showing an example of a neural network according to Embodiment 1 Diagram showing the normal winding state where the wire rope is wound in alignment on the drum Diagram showing a situation where a rope gap has occurred in the wire rope wound around the drum Diagram showing a situation where a wire rope wrapped around a drum runs aground Diagram showing a situation where multiple running aground occurs on a wire rope wrapped around a drum Diagram showing a situation where rope loosening occurs in the wire rope wound around the drum A diagram showing an example of a neural network according to Embodiment 2

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the figures, the same or equivalent parts are given the same reference numerals.

Embodiment 1.
FIG. 1 is a block diagram showing the configuration of a winch drum irregular winding detection device according to an embodiment of the present invention. The winch drum irregular winding detection device 1 detects the irregular winding state when the wire rope 3 wound around the drum 2 of the winch is not aligned. A wire rope 3 wound around the drum 2 extends via a guide sheave and a top sheave (not shown) to a hook for hanging a load. In FIG. 1, the drum 2 is represented in cross section.

The winch drum irregular winding detection device 1 includes an imaging device 10 that photographs the wire rope 3 wound around the drum 2 to generate image data, and a winch drum irregular winding detection device 1 that detects the wire rope wound around the drum 2 from the image data generated by the imaging device 10. 3 is in a disorderly winding state or in a normal winding state. The image of the wire rope 3 taken by the imaging device 10 may be taken at any position on the drum 2, but in order to quickly detect irregular winding, the image should be taken at the winding position 4 on the drum 2 where the wire rope 3 is wound. It is desirable to include. It is desirable that the imaging device 10 be capable of photographing the entire length of the drum 2 in the axial direction so that irregular winding can be detected at any position on the drum 2.

The irregular winding determination section 11 includes an image acquisition section 12 , an image analysis section 13 , a determination section 14 , and a notification section 15 . The image acquisition unit 12 acquires image data generated by the imaging device 10. The imaging device 10 photographs the wire rope 3 wound around the drum 2 and generates, for example, 30 frames of image data per second. A plurality of image data that partially overlap each other and cover the entire circumference of the drum 2 are captured. The image acquisition unit 12 captures image data every time the drum 2 rotates by a fixed angle, for example, 90 degrees, and captures a fixed number of image data, for example, four images, while the drum 2 rotates once.

The imaging device 10 may be a camera that takes still images. In that case, the image acquisition unit 12 sends an imaging command to the imaging device 10 in accordance with the rotation of the drum 2, and acquires the image data at that time from the imaging device 10. The image data acquired by the image acquisition unit 12 does not need to be synchronized with the rotation of the drum 2. For example, when the drum 2 rotates at the maximum rotational speed, a plurality of pieces of image data, some of which overlap each other and cover the entire circumference of the drum 2, are captured at a constant cycle during one rotation of the drum 2. , image data may be acquired.

The image analysis unit 13 includes a trained model of a neural network that analyzes image data, and analyzes the winding state of the wire rope 3 wound around the drum 2. Before analyzing the image data, the image acquisition unit 12 or the image analysis unit 13 normalizes the image data so that the average brightness, maximum brightness, and minimum brightness of each pixel are the same value among the image data. good. The determining unit 14 determines whether the winding is in the irregular winding state or the normal winding state based on the analysis by the image analysis unit 13, and if it is determined that the winding is in the irregular winding state, it notifies that the winding is in the irregular winding state. I will tell this to Department 15.

When the notification unit 15 is informed by the determining unit 14 that the irregular winding is occurring, it notifies the operator that the irregular winding has occurred. For example, the notification unit 15 causes an alarm to sound, a warning light to turn on or blink, or a display device to indicate that a winding state is occurring. Instead of or in addition to the notification, the notification unit 15 may instruct a control device (not shown) to stop the operation of the winch drum.

The random winding determination unit 11 includes an I/O interface, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The CPU is, for example, a microprocessor or the like, and is a central processing unit that executes various processes and calculations. The random winding determining unit 11 reads out a control program stored in the ROM and causes the CPU to execute the control program while using the RAM as a work memory. , and the operation of the notification section 15.

FIG. 2 is a block diagram showing the functional configuration of the image analysis section according to the first embodiment. In the first embodiment, the image analysis section 13 includes a compression neural network 16, a restoration neural network 17, and a comparison section 18.

The compression neural network 16 obtains latent variables from the image data using probability distributions. The latent variables include, for example, the winding diameter of the wire rope 3, the number of windings in one layer of the wire rope 3, the moving direction of the winding position 4 (left to right or right to left), and the winding position of the wire rope 3 (drum the number of layers from the end), the contact angle of the wire rope 3, the diameter of the wire rope 3, the length of the drum 2 (flange spacing), and the number of remaining layers that can be wound up to the flange diameter, or variables that are a combination of these. It is.

The restoration neural network 17 generates reproduction data having the same dimensions as the image data from the latent variables acquired by the compression neural network 16. The comparison unit 18 compares the image data input to the compression neural network 16 and the reproduced data generated by the restoration neural network 17 to detect different parts of the image.

FIG. 3 is a diagram illustrating an example of a neural network according to the first embodiment. The compression neural network 16 and the restoration neural network 17 each have an input layer or an output layer that includes nodes composed of artificial neurons (hereinafter simply referred to as neurons), an intermediate layer, and interconnect nodes between adjacent layers. Consists of edges that connect to. The intermediate layer includes one or more n-layers. Each pixel value x _i of image data is input to each node i (i=1...k) of the input layer. In the middle layer, the outputs of each previous layer are combined and the results calculated by the activation function are transmitted to the next layer, and finally reproduced from each node j (j = 1...k) of the output layer. The pixel value _zj of the data is output.

The compression neural network 16 compresses the features of the image data x ₀ to x _k input to the input layer by repeating the convolution layer and the pooling layer to obtain latent variables y ₁ to y _M. The restoration neural network 17 is a probabilistic model that follows the distribution law, conversely from the latent variables y ₁ to y _M , and uses the data distribution as a probability, and reproduces the reproduced data z ₀ to z of the same dimension as the image data x ₀ to x _k . _k is generated and output from the output layer.

The compression neural network 16 and the restoration neural network 17 use, as learning data, image data generated by photographing the wire rope 3 in a normal winding state where the wire rope 3 is aligned and wound around the drum 2, and calculates the difference between the image data and the reproduced data. Let the machine learn to stay within the specified range. The image data subjected to machine learning includes at least image data of a normal winding state when the wire rope 3 is wound around the drum 2 in two or more layers. It is desirable that the image data for learning includes a normal winding state in which the wire rope 3 is wound around the drum 2 in only one layer.

The comparison unit 18 compares the image data and the reproduced data for each corresponding pixel, and extracts different pixels whose pixel values have a difference equal to or greater than a threshold value. Using the different pixels as a restoration error, the weights of the compression neural network 16 and the restoration neural network 17 are adjusted by error backpropagation.

For example, the compression neural network 16 and the restoration neural network 17 are subjected to machine learning so that the number of different pixels is less than or equal to a specified number and the minimum distance between different pixels is greater than or equal to the specified number. The compression neural network 16 and the restoration neural network 17 are trained models of neural networks.

The image analysis unit 13 generates reproduction data from the image data using the trained model of the neural network shown in FIG. Since the compression neural network 16 and the restoration neural network 17 are machine-learned using image data of a normal winding state, latent variables obtained from image data of a disorderly winding state in which the wire rope 3 wound around the drum 2 is not aligned. However, playback data of a normal winding state in which the wire rope 3 is wound on the drum 2 in alignment is generated. For image data in a normal winding state, the difference between the image data and the reproduced data is less than the specified value, but for image data in an irregular winding state, the reproduced data is in a normal winding state, so the difference between the image data and the reproduced data is less than the specified value. Becomes larger than specified.

The comparison unit 18 compares the image data and the reproduced data for each corresponding pixel, and extracts different pixels whose pixel values have a difference equal to or greater than a threshold value. The comparison unit 18 sends the extracted different pixels to the determination unit 14. The determination unit 14 compares the image data and the reproduced data, and determines that there is an abnormality when there is a difference greater than a threshold value. For example, if the number of different pixels exceeds a predetermined number and is unevenly distributed within a predetermined size range in the images, the determination unit 14 determines that there is a difference between the two images, and determines that there is a difference between the two images. It is determined that the data is in a disordered state.

FIG. 4 is a diagram showing a normal winding state in which the wire rope is wound in alignment on the drum. In a normal winding state, the wire rope 3 is always in contact with the wire rope 3 that was wound immediately before, and the lower wire rope 3 or the side surface of the drum 2 at the winding position 4. The winding position 4 moves sequentially from the left flange to the right flange, for example, and when the winding position 4 reaches the right flange, it rides on the wire rope 3 that was wound just before, and the winding position 4 moves from the left flange to the right flange. Move sequentially from the right flange to the left flange. The wire rope 3 is wound around the drum 2 in multiple layers by repeating the movement of the winding position 4 alternately from left to right.

FIG. 5 is a diagram showing a state where a rope gap has occurred in the wire rope wound around the drum. The rope gap 5 is a state in which a gap is created between adjacent wire ropes 3 without contacting the wire rope 3 that was wound immediately before at the winding position 4. If the winding of the wire rope 3 is continued in a state where the rope gap 5 is created, the wire rope 3 in the upper layer falls into the rope gap 5, and the wire rope 3 is damaged. The rope gap 5 is a kind of disorderly winding state. The comparison unit 18 detects the rope gap 5 as a different pixel.

FIG. 6 is a diagram showing a state in which a wire rope wound around a drum runs over the wire rope. The wire rope 3 running over 6 is a state in which the wire rope 3 rides on the wire rope 3 that was wound immediately before in the middle of the drum 2. The wire rope 3 that runs aground is subjected to a larger force than usual at the location where it runs aground, and is not regulated in the axial direction of the drum 2, causing a rope gap 5 and further running 6. The comparison unit 18 detects the overlapping portion as a different pixel. At the end of the drum 2, it is normal for the wire rope 3 folded back at the flange to ride on the wire rope 3 in contact with the flange, so this is not a case of random winding 6.

FIG. 7 is a diagram showing a state in which multiple running over occurs on the wire rope wound around the drum. The multiple running 7 is a state in which the wire rope 3 that is folded back at the flange and rides on the wire rope 3 that contacts the flange at the end of the drum 2 is further run on. Although the multiple running-up 7 may occur even in the middle of the drum 2, the running-up occurs before the multiple running-up 7. In the multiple stranding 7, the stranded wire rope 3 cannot support the tension of the stranded wire rope 3 and collapses to the lower layer at once, and the wire rope 3 is damaged by the impact. Alternatively, the wire rope 3 may climb over the flange and escape from the rope. The comparison unit 18 detects the multiple overlapped portion 7 as a different pixel.

FIG. 8 is a diagram showing a state in which the wire rope wound around the drum has become loose. The rope slack 8 is a state in which a gap is created between the upper layer and the lower layer of the wire rope 3 wound around the drum 2. This occurs when the wire rope 3 is being sent out to lower the load, and the drum 2 is further rotated in the sending direction after the load has landed. When rope slack 8 occurs, the loosened wire rope 3 has no friction with the layer below and is not regulated in the axial direction of the drum 2, creating a rope gap 5 and winding the wire rope 3 in the state of rope slack 8. If it is removed, a rope gap 5 or a run-on 6 will occur. The comparison unit 18 detects the rope slack portion 8 as a different pixel.

The determination unit 14 may further recognize the arrangement of different pixels and determine the rope gap 5, run-up 6, multiple run-up 7, and rope slack 8, which are each state of irregular winding.

According to the winch drum irregular winding detection device 1 of the first embodiment, the compression neural network 16 and the reproduction neural network are machine-trained to reduce the difference between image data and reproduced data using image data in a normal winding state. If the difference between the image data and the reproduced data exceeds a predetermined range, it is determined that the image data is in a state of random winding, so random winding occurs regardless of the number of layers of the wire rope 3 wound around the drum 2. can be detected. In the first embodiment, since machine learning is performed using image data of a normal winding state, the winch drum irregular winding detection device 1 can be implemented even without image data of a irregular winding state.

Embodiment 2.
FIG. 9 is a diagram illustrating an example of a neural network according to the second embodiment. In the second embodiment as well, the winch drum irregular winding detection device 1 includes an imaging device 10 and an irregular winding determination section 11, as shown in FIG. It includes a section 13, a determining section 14, and a notification section 15. In the second embodiment, the image analysis unit 13 divides the image data into each state of the irregular winding state, for example, each state of rope gap 5, run-up 6, multiple run-up 7, and rope slack 8, and the normal winding state. This is a classification neural network 19 for classifying.

The classification neural network 19 is composed of an input layer, a middle layer, and an output layer each including nodes composed of artificial neurons (hereinafter simply referred to as neurons), and edges interconnecting nodes between adjacent layers. be done. The intermediate layer includes one or more n-layers. Each pixel value xi of image data is input to each node i (i=1...k) of the input layer. In the intermediate layer, the outputs of the previous layers are combined, and the results calculated by the activation function are transmitted to the subsequent layer and finally output to the output layer. In the intermediate layer, by repeating the convolution layer and the pooling layer, the image data is classified into irregular winding states and normal winding states. The output layer has nodes for each state of the random winding state and the number of nodes for the normal winding state, and each node j (j=1...M) outputs the probability pj that image data is classified into state j. .

The classification neural network 19 inputs a plurality of image data classified into each state of the irregular winding state and the normal winding state, and calculates the difference between the classified state of the image data and the output of the classification neural network 19. Perform machine learning by backpropagating and adjusting each parameter. The image data to be subjected to machine learning includes at least image data of irregular winding states and normal winding states when the wire rope 3 is wound around the drum 2 in two or more layers. It is desirable that the image data used for learning includes each state of the irregular winding state and the normal winding state when the wire rope 3 is wound in only one layer on the drum 2.

The randomization determination unit 11 inputs the image data to the trained model of the classification neural network 19, and selects the node with the highest probability of being the output of the classification neural network 19, that is, the most probable state, to which the image data should be classified. It is determined that the condition is The determination unit 14 determines whether the image data is in a normal winding state or a disorderly winding state based on the probability of being classified into each state which is the output of the classification neural network 19 as the image analysis unit 13.

If the determining unit 14 determines that the winding is in a state of irregular winding, it notifies the notifying unit 15 that the winding is in a state of irregular winding. When the notification unit 15 is informed by the determining unit 14 that the irregular winding is occurring, it notifies the operator that the irregular winding has occurred. For example, the notification unit 15 causes an alarm to sound, a warning light to turn on or blink, or a display device to indicate that a winding state is occurring. Instead of or in addition to the notification, the notification unit 15 may instruct a control device (not shown) to stop the operation of the winch drum.

The classification neural network 19 may classify the normal winding state into a plurality of states. For example, the winding position 4 of the wire rope 3 is at the left end of the drum 2, the winding position 4 is in the middle of the drum 2 and moving from left to right, and the winding position 4 is in the middle of the drum 2 and moving from left to right. The normal winding state can be classified into a state in which the winding position 4 is moved to the left from the position 4, and a state in which the winding position 4 is at the right end of the drum 2.

According to the winch drum irregular winding detection device 1 of the second embodiment, image data is classified using a trained model of a neural network that is machine-trained using image data of each state of irregular winding state and normal winding state. , the occurrence of random winding can be detected regardless of the number of layers of the wire rope 3 wound around the drum 2. In the second embodiment, it is further possible to classify and detect which state is the irregular winding state.

1 Winch drum irregular winding detection device 2 Drum 3 Wire rope 4 Winding position 5 Rope gap 6 Running over 7 Multiple running over 8 Rope loosening 10 Imaging device 11 Uneven winding determination section 12 Image acquisition section 13 Image analysis section 14 Judgment section 15 Notification section 16 Compression neural network 17 Restoration neural network 18 Comparison section 19 Classification neural network

Claims (2)

an imaging device that generates image data by photographing a wire rope wrapped around a drum;
The wire rope includes a trained model of a neural network that analyzes image data generated by the imaging device, and the wire rope is in a normal winding state where the wire rope is wound in alignment on the drum, or the wire rope is wound around the drum. a random winding determination unit that determines whether the winding is in a random winding state where the windings are not aligned;
Equipped with
The random winding determination unit is
a compression neural network that obtains a latent variable of the state in which the wire rope is wound around the drum from the image data;
a restoration neural network that generates reproduced data of the same dimension as the image data from the latent variables acquired by the compression neural network;
a comparison unit that compares the image data input to the compression neural network and the reproduction data generated by the restoration neural network to detect different parts of the image;
including;
The compression neural network and the restoration neural network use, as learning data, image data generated by photographing the wire rope in a normal winding state where the wire rope is wound in alignment on the drum, and combine the image data and the reproduction data. It is a trained model of a neural network that is machine-trained to keep the difference within a specified range.
The irregular winding determination unit determines that the image data is in the irregular winding state when a difference between the image data and the reproduced data generated by the learned model exceeds a predetermined range.
Winch drum irregular winding detection device. an imaging device that generates image data by photographing a wire rope wrapped around a drum;
The wire rope includes a trained model of a neural network that analyzes image data generated by the imaging device, and the wire rope is in a normal winding state where the wire rope is wound in alignment on the drum, or the wire rope is wound around the drum. a random winding determination unit that determines whether the winding is in a random winding state where the windings are not aligned;
Equipped with
The random winding determination unit is
A neural network uses image data of a normal winding state in which the wire rope is wound in alignment on the drum, and image data of a disorderly winding state in which the wire rope wound around the drum is not aligned, as learning data. a trained model of a neural network that is subjected to machine learning to classify the image data into the normal winding state and the irregular winding state;
The trained model of the neural network classifies the image data generated by the imaging device into each state of irregular winding including rope gap, run-up, multiple run-up, and rope slack, and the normal winding state. Outputs the probability that
The irregular winding determination unit determines whether the image data is in the normal winding state or in the irregular winding state, based on the probability of being classified into each category output by the learned model.
Winch drum irregular winding detection device.

Images