JP2023013842A - Chain elongation detection device - Google Patents

Abstract

To provide a chain elongation detection device capable of accurately detecting partial elongation of a chain.SOLUTION: A chain elongation detection device includes: a measuring unit 15 for continuously measuring a total elongation amount of a plurality of links in a predetermined section of a chain 10 that meshes with a pair of sprockets 12, 13; and an estimation unit 22 for estimating an elongation amount per link by inputting the total elongation amount measured by the measuring unit 15 into a system that is trained with a number of data sets that combine elongation amount data for each link of one lap of a chain and data corresponding to the total elongation amount measured by the measuring unit 15.SELECTED DRAWING: Figure 1

Description

An embodiment of the present invention relates to a chain elongation detection device capable of detecting partial elongation of a chain.

In general, conveying devices such as escalators and auto roads are provided with steps on which passengers and objects are placed, and moving handrails that are held by passengers. do. As is well known, a chain is constructed by arranging shafts composed of pins, bushes, and rollers at a predetermined pitch, and alternately connecting these shafts with inner links and outer links.

A chain having such a structure elongates over time due to the above-described wear of the shaft portion and the like. If this elongation becomes large, it may not mesh well with the sprocket, causing the teeth to jump over the sprocket teeth. To prevent such problems from occurring, it has been proposed to measure and monitor the elongation of the chain (see, for example, Patent Document 1).

The technique for measuring the elongation of the chain disclosed in Patent Document 1 is to install photoelectric sensors at two mutually separated locations along the running direction of the chain to be measured, and transmit the signals to a microcomputer for measurement. connected to. The distance between the two photoelectric sensors is set to N times the official pitch of the chain. Each photoelectric sensor is installed so that the optical axis is blocked when each shaft of the chain passes through the optical axis. When the chain is not stretched, the rollers pass through the optical axes of the two photoelectric sensors at the same time, so the sensor signals are turned on at the same time. However, as the chain lengthens due to long-term use, a time lag occurs in the ON time of the sensor signal in proportion to the length of the chain. This time difference is proportional to the amount of chain elongation, and the total elongation of N consecutive links located between the two photoelectric sensors is measured. The chain speed is assumed to be constant during the measurement.

Between the two photoelectric sensors, there are always N (for example, 15) links as the chain runs, and the total elongation of these N links is continuously measured. Since the chain stretches essentially uniformly, this measurement technique is usually sufficient to monitor the amount of chain elongation over time and determine if the chain needs to be replaced.

However, in reality, although the frequency of occurrence is low, partial elongation occurs in which only one link is greatly elongated, and the chain needs to be replaced. In this case, the above-mentioned measurement technology measures the total amount of elongation of N links that enter a predetermined range (between two photoelectric sensors) as the chain runs. It will be buried in the chain and will be judged as an elongation amount that does not reach the threshold for chain replacement. Therefore, there is a high possibility that the situation that the chain needs to be replaced due to partial elongation cannot be correctly determined and is overlooked. In that case, in the worst case, the chain link at that portion may break.

Japanese Patent No. 6505890

Conventional measurement technology cannot detect random partial elongation of one link in one lap of the chain, a small number of links below a predetermined number, or links located far apart. rice field.
SUMMARY OF THE INVENTION It is an object of the present invention to provide a chain elongation detection device capable of accurately detecting partial elongation of a chain.

A chain elongation detection device according to an embodiment of the present invention detects the elongation of a chain that is spanned between a pair of sprockets with a plurality of links connected in an endless manner. A measuring unit that continuously measures the total elongation of a plurality of links in a predetermined section of a chain that meshes and runs in the longitudinal direction, elongation data for each link for one circumference of the chain, and the measuring unit. estimating the elongation for each link by inputting the total elongation measured by the measuring unit into a system trained with multiple data sets combined with data corresponding to the total elongation measured by and an estimating unit that

According to the above configuration, it is possible to accurately detect partial elongation of the chain, which has been difficult to detect in the past, and to reliably replace the chain in which partial elongation has occurred.

1 is a configuration diagram of a chain elongation detection device according to an embodiment of the present invention; FIG. 1 is an exploded perspective view of a chain whose elongation is to be detected in one embodiment; FIG. It is a figure explaining an example of sensor arrangement of a measuring part of chain elongation in one embodiment. (a) is a diagram showing the elongation of each link of the chain measured by the chain elongation measuring section, and (b) is a waveform diagram showing the amount of elongation of each link for one circumference of the chain. (a) is a diagram showing the measurement state of elongation of a plurality of links in a predetermined section determined by the installation position of the sensor, (b) is the partial elongation shown in FIG. 3 (b) in the measurement state shown in (a). FIG. 10 is a waveform diagram showing changes in the total elongation amount of a plurality of links when the elongation of the chain is measured. FIG. 4 is a diagram showing a deep learning model that is an example of a chain elongation estimator in one embodiment. 6 shows an example of a training data set for the deep learning model shown in FIG. 5, where (a) shows target data on the output side and (b) shows data on the input side. FIG. 6 shows an example of a whole number of training data sets for the deep learning model shown in FIG. 5, where (a) shows target data on the output side and (b) shows data on the input side. 6 shows another example of many training data sets for the deep learning model shown in FIG. 5, where (a) shows target data on the output side and (b) shows data on the input side. 6 shows still another example of a large number of training data sets for the deep learning model shown in FIG. 5, where (a) shows target data on the output side and (b) shows data on the input side. FIG. 4 is a configuration diagram showing an example in which a plurality of estimating units are provided in the chain elongation detection device according to the embodiment; FIG. 4 is a configuration diagram showing an example in which an estimation unit in the chain elongation detection device according to the embodiment is provided on the escalator microcomputer side; FIG. 7 is a diagram illustrating a chain elongation measuring unit in a chain elongation detection device according to another embodiment of the present invention; (a) is a diagram showing a state in which only one link is greatly stretched during one lap of the chain, and (b) is the sum of multiple links in a predetermined measurement section, measured by the elongation measurement section of FIG. 12 for the chain in (a). It is a wave form diagram showing the measurement result of elongation amount. When the actual measurement result of the partially stretched chain shown in FIG. 13 is shifted and used as a training data set 1 for the deep learning model, (a) shows the partially stretched waveform that is the target data on the output side, and (b) shows the input FIG. 10 is a diagram showing measured waveforms on the side. When the actual measurement result of the partially stretched chain shown in FIG. 13 is shifted and used as a training data set 2 for the deep learning model, (a) shows the partially stretched waveform that is the target data on the output side, and (b) shows the input FIG. 10 is a diagram showing measured waveforms on the side. When the actual measurement result of the partially stretched chain shown in FIG. 13 is shifted and used as a training data set 3 for the deep learning model, (a) shows the partially stretched waveform that is the target data on the output side, and (b) shows the input FIG. 10 is a diagram showing measured waveforms on the side. When the actual measurement result of the partially stretched chain shown in FIG. 13 is shifted and used as a training data set 4 for the deep learning model, (a) shows the partially stretched waveform that is the target data on the output side, and (b) shows the input FIG. 10 is a diagram showing measured waveforms on the side.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows a chain elongation detection device according to an embodiment of the present invention. FIG. 1 shows the drive chain portion of an escalator. Here, an escalator includes a drive chain, a handrail drive chain, a step chain, and the like, but in this embodiment, the drive chain is exemplified as the chain 10 whose elongation is to be detected.

The chain 10 is formed endlessly and is stretched between a pair of sprockets (a driving sprocket 12 and a driven sprocket 13). A drive sprocket 12 is provided on a speed reducer 11 of an escalator drive motor. The driven sprocket 13 is used to drive endless steps and handrail belts (not shown). Then, it is driven by meshing with the pair of sprockets 12 and 13 and travels around along the length direction.

As shown in FIG. 2A, the chain 10 has shafts 101 arranged at a predetermined pitch, and links 102 connect the ends of the adjacent shafts 101 . In addition, the shaft portion 101 is composed of a pin 101C, a bush 101b, and a roller 101a. Further, the link 102 has an outer link 102a and an inner link 102b, and these outer links 102a and inner links 102b alternately connect the shaft portions 101 together.

Returning to FIG. 1, the chain elongation detection device according to this embodiment includes a chain elongation measuring unit 15, a transmitting device 20 for transmitting measurement data measured by the measuring unit 15 to a monitoring center 21, and a monitoring center 21. and an estimation unit 22 for estimating the elongation amount of each link of the chain 10 from the transmitted measurement data.

The chain elongation measuring unit 15 is basically the same as that described in the prior art, and includes two sensors (hereinafter referred to as photoelectric sensors) 151 and 152 and an elongation amount calculator configured in the escalator microcomputer 19. and a portion 153 . As shown in FIG. 2B, the chain elongation measuring unit 15 measures the total elongation of the plurality of links 102 of the chain 10 running in a predetermined section defined by the installation positions of the two photoelectric sensors 151 and 152. Measure continuously. Note that FIG. 2B shows only the outer link of the link 102 and omits the illustration of the inner link.

The two photoelectric sensors 151 and 152 are arranged along the direction of movement of the chain 10 at intervals of an integer multiple of the pitch of the chain 10 . For example, as shown in FIGS. 1 and 2B, the first and second photoelectric sensors 151 and 152 are arranged at left and right positions avoiding the chain break detection component 18 arranged on the chain 10 . The installation interval between the photoelectric sensors 151 and 152, that is, the distance between the optical axes, is the pitch interval of the links of the chain 10, for example, 15×pitch.

As shown in FIG. 2B, the photoelectric sensors 151 and 152 are composed of a light projecting portion and a light receiving portion, and each time the shaft portion 101 of the chain 10 passes through the optical axis between them, a signal is output to the elongation amount calculating portion 153. do. The elongation amount calculator 153 calculates the elongation amount of the chain 10 based on the magnitude of the time lag between the two detection signals input from the two photoelectric sensors 151 and 152 .

The estimation unit 22 estimates the amount of elongation of each link of the chain 10 from the measurement data transmitted from the measurement unit 15, as described above. The estimating unit 22 has an estimating system, such as a deep learning model, trained using a large number of data sets, and uses this system to estimate the amount of elongation of each link of the chain 10 . The training data set is a combination of elongation amount data for each link of the chain 10 for one round and data corresponding to the total elongation amount measured by the measuring unit 15. 1000 or 2000) are prepared and trained. That is, training is performed with data corresponding to the above-mentioned total elongation amount as input and elongation amount data for each link as a target. By inputting the total elongation measured by the measuring unit 15 into the system trained in this way, the elongation for each link is estimated.

FIG. 3(a) shows an elongation measurement portion of one link of the chain, and FIG. are shown for each. As described above, the links 102 of the chain 10 include outer links 102a and inner links 102b, which alternately connect the shafts 101 together. As is well known, there is approximately zero stretch between inner links 102b and stretch between outer links 102a. Therefore, as shown in FIG. 3(b), the amount of elongation in a saw-tooth shape in which the non-elongation and the elongation are alternately shown.

As shown in FIG. 3(b), each link 102 of the chain 10 has a substantially equal amount of elongation, but there are rare cases where only one link is greatly elongated as shown in the figure. If the portion that extends by one link exceeds the replacement value as shown in the figure, the chain needs to be replaced. FIG. 4 shows measured values of the elongation amount of the chain 10 of FIG. 3 measured by the measuring unit 15 shown in FIG. The photoelectric sensors 151 and 152 are installed along the moving direction of the chain 10 at intervals of an integer multiple (in this embodiment, 15 times as described above) of the pitch of the chain 10 . As the vehicle travels, as shown in FIG. 4A, the total elongation of 15 consecutive links in the section defined by the installation positions of the photoelectric sensors 151 and 152 is sequentially measured. For this reason, as shown in FIG. 4(b), fine fluctuations in elongation amount for each link shown in FIG. 3 are smoothed and measured, and relatively flat measured values are obtained. In addition, since the measured values for the portion that has been greatly elongated by one link are smoothed, the measured values tend to be buried in the overall measured values, and it is judged that the amount of elongation is smaller than that of the line requiring replacement of the chain.

The estimation unit 22 provided in the monitoring center 21 in FIG. 1 is the smoothed total elongation amount for 15 links shown in FIG. From the measured data, the amount of elongation for each link is estimated. A trained deep learning model shown in FIG. 5 is used as the estimation system of the estimation unit 22 . This deep learning model is configured as a three-layer neural network, and the input side has nodes corresponding to the number of links for one round of the chain 10 (for example, 112 nodes). The input data is the chain elongation measurement data measured by the measuring unit 15, that is, the total elongation amount of 15 consecutive links is measured by shifting the measurement range by one link per circumference of the chain as the chain 10 runs. measurement data.

The measured data input to the trained deep learning model is output through the first layer (336 nodes), the second layer (224 nodes), and the third layer (112 nodes). 112 nodes corresponding to the number of links for one round of the same chain 10 as the side output the estimated value for one round of the chain converted into the amount of elongation for each link.The neural network that becomes this deep learning model is: It is pre-trained to have the above-mentioned inference function by a training data set.

An example of a training data set will be described with reference to FIGS. 6(a) and 6(b). FIG. 6(a) shows simulated elongation data that defines the elongation amount of each link for one cycle of the chain, and this becomes training data (target data: data to be obtained) as an output. This simulated elongation data simulates the actual elongation of the chain, and is a serrated elongation amount in which the non-elongated chain and the elongated chain are alternately repeated. For stretched links, the amount of stretch was set as a random value. That is, for the simulated elongation data, random values were set assuming that the elongation amount of the inner link, which generally does not occur, is 0, and that each of the outer links has an elongation amount that is random.

FIG. 6(b) is a simulation obtained by calculating how the simulated elongation amount of the chain shown in FIG. 6(a) is measured when the measuring unit 15 of FIG. It is measurement data and becomes data on the input side of training data. A large number of data sets, for example, 1000, are prepared by combining the simulated measurement data on the input side and the simulated elongation data on the output side shown in FIG.

An example of this training data set is shown in FIGS. 7(a) and 7(b). A number of these 1st to Mth (1000th in this case) data sets are used to train a deep learning model. If the training can be carried out so that the error is sufficiently small, the measurement data for each 15 links measured by the measurement unit 15 is supplied as described above, and the elongation amount for each link is output with high accuracy.

In the above configuration, the elongation amount of the chain 10 is measured, for example, once a day by the chain elongation detection device described with reference to FIG. That is, the pair of sprockets 12 and 13 are rotationally driven, and the chain 10 is made to circulate in the length direction. As a result of this running, the total elongation of the 15 links within the section defined by the installation positions of the photoelectric sensors 151 and 152 is calculated by the elongation calculator 153 of the escalator microcomputer 19 . The total elongation of every 15 consecutive links of the chain 10 is measured for one round. The measurement data is temporarily stored in the memory within the escalator microcomputer 19 .

This measurement data is transmitted to the monitoring center 21 by the measurement data transmission device 20 at a predetermined time. In the monitoring center 21 , this measurement data is input to a trained deep model serving as the estimation unit 22 . The trained deep model shown in FIG. 5 converts the input measured data for every 15 links into measured data (estimated value) of the elongation amount for each link. This converted data is used to determine whether the chain needs to be replaced.

In this manner, measurement data (estimated value) of the amount of elongation for each link of the chain 10 can be obtained, so even if only one link elongates, this can be accurately detected. That is, in the conventional technology, it is determined whether or not the chain needs to be replaced from the waveform obtained in the form of FIG. 4(b) for the chain elongation amount data. Therefore, when the actual elongation amount of the chain is greatly elongated by one link as shown in FIG. 3, it may be determined that the chain does not need to be replaced.

On the other hand, in this embodiment, the measurement data in FIG. 4 is converted into the measurement value for each link in FIG. Parts can also be detected correctly, and there is no misjudgment. Therefore, it is possible to correctly detect a situation in which only one link is partially stretched, and it becomes possible to automatically instruct an inspector to carry out an on-site inspection or to replace the chain, thereby providing greater safety.

Other examples of training data sets for the estimation unit 22 are shown in FIGS. 8A and 8B and FIGS. 9A and 9B.

FIG. 8 shows a case where 1000 training data sets are prepared as training data mainly assuming partial elongation in which a part of the chain 10 is elongated. FIG. 1(a) shows simulated elongation data indicating elongation of each link of the chain 10, which is output side data (target). FIG. 5(b) is simulated measurement data simulating measurement data for every 15 links measured by the measurement unit 15, calculated based on the elongation amount shown in FIG. training data.

In the simulated elongation data of FIG. 8(a), the second from the top is assumed that the elongation amount of the inner link that does not elong is 0, and that only one outer link is elongated at random during one circumference of the chain. In addition, in the first, third, and 1000th links from the top of FIG. 8(a), the amount of elongation for the inner links that do not extend is 0, and the outer link side has only one location in one circumference of the chain, and a maximum of about five links. Assuming that the following number of consecutive outer links are partially extended, the amount of extension of each link was set as random.

The training data set of FIG. 8 described above is based on the assumption that partial elongation is concentrated at one point in one round of the chain. The amount of elongation of the partially elongated portion in one round of the chain is set to a random value, and the number of the link where the elongation occurs is also set to be random. In addition, the number of partially stretched links generated at one location is also a random number within a small range of 5 links or less. A deep model trained on such a data set can accurately detect various forms of partial elongation in the chain.

In FIG. 8, it is assumed that the partial elongation of one or a few (five or less) links occurs at one point on the circumference of the chain. Assuming that the above-mentioned partial elongation occurs in , the simulated elongation data of FIG. 8(a) may be formed, and based on this, the simulated measurement data of FIG. 8(b) may be calculated by theoretical calculation.

When a deep learning model trained using the training data shown in FIG. 8 is used, training of the deep learning model is facilitated, resulting in an improvement in training accuracy. That is, when training with the training data of FIGS. 6 and 7, the amount of information that the deep learning model must estimate is very large because of the training data in which all the links for one round of the chain are randomly extended. grow to Therefore, it is quite difficult to obtain a model with high estimation accuracy, and obtaining a model with high accuracy is not necessarily guaranteed.

On the other hand, when a deep model trained using the training data shown in FIG. 8 is used, the number of links to be estimated is at most about 5 links in one round of the chain, and the amount of information estimated by the deep model is greatly reduced. Therefore, training of the deep learning model is facilitated, and it can be expected that the estimation accuracy of the obtained model will be high, and the extension amount for each link can be estimated with higher accuracy. However, for chain data that looks like it has been stretched as a whole, there is a possibility that the estimation accuracy will decrease because it differs from the configuration of the training model.

The training data shown in FIG. 9 assumes that the simulated elongation data (a) in FIG. 9 assumes that the elongation amounts of all the outer links are all uniform constant values. That is, in this case, it is assumed that the chain as a whole is uniformly stretched rather than partially stretched. The stretched side and the non-stretched side appeared alternately, but the amount of stretch was assumed to be the same for each circumference of the chain. This is similar to the most common elongation in an actual chain.

In the training data shown in FIG. 9, the elongation values of all links were constant. Therefore, compared to the case where the elongation amounts of all the links in FIGS. 6 and 7 are random, the training becomes easier, and the estimation accuracy after the training becomes higher. In addition, since FIG. 9 assumes a general chain elongation that occurs most frequently, the effects of sensor noise and other factors mixed in the measurement data are also considered using the training data in FIG. The sensor noise is removed from the result of giving the measured data to the trained deep model and outputting it. Therefore, it can be expected that the average elongation amount of the chain can be detected more accurately than the conventional technique that does not pass the deep model.

Each of the training data described above is normalized so that the maximum value of the simulated measurement data on the input side is 1 or a specific numerical value of 1 or less, except when all data are 0. .
Also, the training data output side may include not only simulated elongation data generated in a simulation, but also actual chain elongation data actually measured by the measurement system.

In the chain elongation measuring device of FIG. 10, a monitoring center 21 is provided with a plurality of trained deep learning models (22A, 22B), which are estimation units 22, and their output values are received to comprehensively determine the state of the chain. A comprehensive determination unit 23 is provided. In this case, the deep learning models 22A and 22B are assumed to have different characteristics. For example, deep learning model 22A should be trained with the training data of FIG. 8, and deep learning model 22B should be trained with the training data of FIG. When configured in this way, the deep learning model 22A mainly extracts a partially large stretched portion, and the deep learning model 22B extracts an average stretch amount. The comprehensive judgment unit 23 judges whether chain replacement is necessary by looking at the extraction results of both.

With the chain elongation measuring apparatus of FIG. 10, both the case where the average amount of elongation is large and the case where the chain is partially elongated can be reliably detected. Therefore, it is possible to ensure the safety of passengers and perform maintenance with higher accuracy without overlooking elongation of the chain and abnormalities.

In the chain elongation measuring device of FIG. 11, the trained deep learning model, which is the estimation unit 22, is configured in the escalator microcomputer 19 instead of the monitoring center 21 side. With this configuration, the elongation amount for each link is estimated inside the escalator microcomputer 19 . For this reason, on the side of each escalator, estimated data on the amount of elongation for each link is extracted and output to the monitoring center 21 side. Since the monitoring center 21 side does not process the data sent from each escalator one by one, it is only necessary to judge whether there is an abnormality in the chain or whether the chain needs to be replaced, thus reducing the processing on the monitoring center 21 side. has the effect of being

In any of the above-described embodiments, the two sensors 151 and 152 for detecting elongation of the chain are photoelectric sensors, but they are not necessarily limited to photoelectric sensors. It doesn't matter what kind. For example, it applies to all systems that measure the total elongation of continuous chain sections, such as laser displacement sensors and proximity sensors.

In the above-described embodiment, the chain elongation measuring unit 15 shown in FIGS. 1, 10 and 11 has a plurality of links (15 The total elongation of the 15 links between the sensors 151 and 152 is detected as the chain 10 moves. A measuring unit may be used. In the following embodiments, a chain elongation measuring section 25 configured as shown in FIG. 12 is used.

A chain 10 to be measured is stretched between a pair of sprockets 12 and 13 . A first sensor 251 and a second sensor 252 that constitute the chain elongation measuring unit 25 are arranged near the corresponding sprockets 12 and 13 to detect the timing at which the teeth of the sprockets 12 and 13 pass and output a detection signal. do. The elongation of the chain 10 is measured based on the timing of generation of these detection signals. This measurement principle has been filed as Japanese Patent Application No. 2020-131993 by the applicant of the present application. FIG. 12 illustrates the principle of measurement, and the sprockets 12 and 13 are drawn to have the same diameter. It is the same even if there is.

The chain 10 is driven by the rotation of the sprockets 12 and 13 and circulates in its length direction. At this time, the first sensor 251 detects the passage timing of the teeth of the sprocket 12 and outputs a first detection signal. The second sensor 252 detects the passage timing of the teeth of the sprocket 13 and outputs a second detection signal. These first and second detection signals are input to a calculation section 253 for calculating chain elongation, and the elongation of the chain 10 is measured based on the phase difference between these first and second detection signals.

The calculation unit 253 corresponds to the elongation amount calculation unit 153 shown in FIG. 1 and the like, and the calculation result is sent to the monitoring center 21 by the transmission device 20 shown in FIG. As the first sensor 251 and the second sensor 252, a transmissive photoelectric sensor, a reflective photoelectric sensor, a proximity sensor, or the like can be used.

The operation of this chain elongation measuring section 25 will be described. As shown in FIG. 13(a), it is assumed that only one link in one lap of the chain 10 is greatly elongated. As the chain 10 moves, the measured value of the measuring unit 25, that is, the first sensor 251, is measured from the starting point A of meshing with the sprocket 13 shown in FIG. And the amount of elongation detected by the second sensor 252 and calculated by the calculator 253 changes as shown in FIG. 13(b). This elongation amount is the total elongation amount of the number of links in the predetermined section from the start point A to the end point B described above.

The measured elongation amount is input to the estimation unit 22 shown in FIG. 1, and the elongation amount for each link is estimated by the deep learning model shown in FIG. In this case, training of the deep learning model that constitutes the estimation unit 22 is performed using the training data sets shown in FIGS. 14 to 17 . The training data sets shown in FIGS. 14 to 17 are obtained by shifting the link positions of the measured data shown in FIG. 13, and a large number of these are prepared and used as training data.

This is because the measurement result measured by the measuring unit 25 for the elongation of only one link shown in FIG. 13(a) is non-linear as shown in FIG. 13(b). This is because it is difficult to theoretically create the total elongation of a plurality of links (input side) based on the elongation of each link (output side) in (a), as in the data set of the embodiment shown above. .

Thus, since the measured waveform cannot be calculated theoretically, a chain that has been stretched by one link in one lap is mounted, and the amount of stretch is actually measured by the chain stretch measuring unit 25 in FIG. The measured waveform indicated by is used as a reference. By using the measurement data of this one test, a measurement waveform is formed in which the position of one link elongation is shifted as shown in FIGS. 14 to 17 . Such measured waveforms can be easily obtained by data processing, so a single piece of measured data is used to prepare a large number of training data sets and train a deep learning model.

By inputting the total value of a plurality of elongations measured by the measuring unit 25 to the estimating unit 22 trained in this manner, the partial elongation of the chain 10 can be accurately detected in the same manner as in the above-described embodiment. can.

While several embodiments of the invention have been described, these embodiments have been presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 10... Chain 101... Shaft part 102... Link 11... Reduction gear 12, 13... Pair of sprockets 15, 25... Elongation measurement part 151, 152, 251, 252... Sensor 153, 253... Elongation amount calculation part 18... Chain DISCONNECTION DETECTION DEVICE 19 Escalator microcomputer 20 Measured data transmission device 21 Monitoring center 22 Guessing part 23 Comprehensive judging part

This measurement data is transmitted to the monitoring center 21 by the measurement data transmission device 20 at a predetermined time. In the monitoring center 21 , this measured data is input to a trained deep learning model that serves as the estimator 22 . The trained deep learning model shown in FIG. 5 converts the input measurement data for every 15 links into measurement data (estimated value) of elongation amount for each link. This converted data is used to determine whether the chain needs to be replaced.

On the other hand, in this embodiment, the measurement data in FIG. 4 is converted into the measurement value for each link in FIG. Parts can also be detected correctly, and there is no misjudgment. Therefore, it is possible to correctly detect a situation in which only one link is partially stretched, and it becomes possible to automatically instruct an inspector to carry out an on-site inspection or to replace the chain, thereby providing greater safety.

The training data set of FIG. 8 described above is based on the assumption that partial elongation is concentrated at one point in one round of the chain. The amount of elongation of the partially elongated portion in one round of the chain is set to a random value, and the number of the link where the elongation occurs is also set to be random. In addition, the number of partially stretched links generated at one location is also a random number within a small range of 5 links or less. A deep learning model trained on such a dataset can accurately detect various forms of partial elongation in the chain.

On the other hand, when using the deep learning model trained with the training data shown in FIG. 8, the part to be estimated is at most about 5 links in one round of the chain, and the information estimated by the deep learning model is significantly less. Become. Therefore, training of the deep learning model is facilitated, and it can be expected that the estimation accuracy of the obtained model will be improved, and the extension amount for each link can be estimated with higher accuracy. However, for chain data that looks like it has been stretched as a whole, there is a possibility that the estimation accuracy will decrease because it differs from the configuration of the training model.

In the training data shown in FIG. 9, the elongation values of all links were constant. Therefore, compared to the case where the elongation amounts of all the links in FIGS. 6 and 7 are random, the training becomes easier, and the estimation accuracy after the training becomes higher. In addition, since FIG. 9 assumes a general chain elongation that occurs most frequently, the effects of sensor noise and other factors mixed in the measurement data are also considered using the training data in FIG. The sensor noise is removed from the output result of giving the measured data to the trained deep learning model. Therefore, it can be expected that the average elongation amount of the chain can be detected more accurately than the conventional technique that does not pass through the deep learning model.

In the chain elongation detection device of FIG. 10, a monitoring center 21 is provided with a plurality of trained deep learning models (22A, 22B), which are estimation units 22, and their output values are received to comprehensively determine the state of the chain. A comprehensive determination unit 23 is provided. In this case, the deep learning models 22A and 22B are assumed to have different characteristics. For example, deep learning model 22A should be trained with the training data of FIG. 8, and deep learning model 22B should be trained with the training data of FIG. When configured in this way, the deep learning model 22A mainly extracts a partially large stretched portion, and the deep learning model 22B extracts an average stretch amount. The comprehensive judgment unit 23 judges whether chain replacement is necessary by looking at the extraction results of both.

With the chain elongation detection device of FIG. 10, both the case where the average amount of elongation is large and the case where the chain is partially elongated to a large extent can be reliably detected. Therefore, it is possible to ensure the safety of passengers and perform maintenance with higher accuracy without overlooking elongation of the chain and abnormalities.

In the chain elongation detection device of FIG. 11, the trained deep learning model, which is the estimation unit 22, is configured in the escalator microcomputer 19 instead of the monitoring center 21 side. With this configuration, the elongation amount for each link is estimated inside the escalator microcomputer 19 . For this reason, on the side of each escalator, estimated data on the amount of elongation for each link is extracted and output to the monitoring center 21 side. Since the monitoring center 21 side does not process the data sent from each escalator one by one, it is only necessary to judge whether there is an abnormality in the chain or whether the chain needs to be replaced, thus reducing the processing on the monitoring center 21 side. has the effect of being

Claims (9)

A chain elongation detection device for detecting elongation of a chain, in which a plurality of links are connected endlessly and spanned between a pair of sprockets,
a measuring unit that continuously measures the total elongation of a plurality of links running in a predetermined section of the chain that meshes with the pair of sprockets and runs in the length direction;
A system trained with a large number of data sets combining elongation amount data for each link of one circumference of the chain and data corresponding to the total elongation amount measured by the measurement unit is provided with the data measured by the measurement unit. an estimation unit for estimating an elongation amount for each link by inputting the total elongation amount;
chain elongation detection device. The data set for training of the estimating unit is the elongation amount data for each link for one circumference of the chain in which partial elongation has occurred in advance, and the starting point of meshing with one of the sprockets of the chain measured by the measuring unit. Using a data set that combines the measured values of the total elongation of a plurality of links in the section up to the end point, a large number of data sets formed by shifting the link positions of these data sets were generated and used as a training data set. The chain elongation detection device according to claim 1, characterized in that: The data set for training of the estimation unit includes simulated elongation data that simulates the amount of elongation of each link for one round of the chain, and multiple links measured by the measuring unit that are calculated by theoretical calculation based on this simulated elongation data. 2. The chain elongation detecting device according to claim 1, wherein the chain elongation detecting device is composed of simulated measurement data corresponding to the total elongation amount of the chain. 4. A random value is set for the simulated elongation data, assuming that an inner link that hardly elongates has an elongation amount of 0, and each outer link has an elongation amount that is random. chain elongation detection device. In the simulated elongation data, the elongation amount of the inner link, which hardly causes elongation, is assumed to be 0. As for the outer link, one link or several distant links in one circumference of the chain are randomly elongated. 4. A chain elongation detecting device according to claim 3, characterized in that the chain elongation detecting device is set to be a vertical one. In the simulated elongation data, the elongation amount of the inner link, which hardly causes any elongation, is assumed to be 0. Regarding the outer links, it is assumed that a maximum of 5 consecutive outer links are elongating, and the elongation amount of each link is random. 6. The chain elongation detecting device according to claim 5, wherein the chain elongation detecting device is set to . 4. The chain elongation detecting device according to claim 3, wherein the simulated elongation data is set such that the elongation amount of the inner link which hardly elongates is 0, and the elongation amount of the outer link is set to a uniform constant value. 4. The chain elongation detection device according to claim 3, wherein the simulated measurement data is normalized so that the maximum value is 1 or a specific numerical value of 1 or less. 4. The chain elongation detecting device according to claim 3, wherein actually measured elongation data is included as elongation amount data for each link for one circumference of the chain in the training data set.

Images