Predictive maintenance method of device through clustered waveforms

Abstract

Disclosed is a predictive maintenance method of a device through clustered waveforms, comprising: an information collection step for collecting information on changes in energy magnitude over time measured by the device and collecting a large amount of energy waveform information for various operations performed by the device; a cluster waveform extraction step for classifying the large amount of energy waveforms collected in the information collection step into similar waveforms to form a plurality of similar clusters, and extracting cluster waveforms from each of the plurality of similar clusters; and a detection step for measuring and collecting real-time energy waveforms from the device when the device is performing an operation in real time, and comparing the collected real-time energy waveforms with information of the plurality of cluster waveforms extracted in the cluster waveform extraction step to detect abnormalies in the device in real time.

Description

Predictive maintenance method for devices using cluster waveforms

[0001] The present invention relates to a predictive maintenance method for a device using a cluster waveform, and more specifically, to a predictive maintenance method for a device using a cluster waveform, which collects a large amount of energy waveform information measured from a device freely performing various operations while performing each operation, classifies the large amount of collected energy waveforms by similar waveforms to form a small number of similar clusters, extracts a small number of cluster waveforms representing each similar cluster based on the similar waveforms belonging to the similar clusters, and then matches and compares real-time energy waveforms measured from a device freely operating in real time with the small number of cluster waveforms to detect an abnormality in the device and issues an alarm to induce management such as replacement or repair before a device malfunctions, thereby preventing economic losses, safety accidents, inconveniences, etc., caused by sudden device malfunctions in advance.

[0002] In general, stable operation is very important for various devices used for automation processes of facilities.

[0003] For example, in the facilities of a large-scale production plant, tens or hundreds of devices are installed and operate in conjunction with each other to continuously produce products. If a single device among the many devices breaks down, a catastrophic situation can occur where the entire operation of the facility comes to a halt.

[0004] At this time, downtime due to equipment failure inevitably leads to enormous losses not only in equipment repair costs but also in operating costs and business effects wasted while the equipment is down.

[0005] To prevent such damage, predictive maintenance methods are being developed to detect and respond to abnormal signs in devices in advance and prevent damage caused by sudden device failures. However, these predictive maintenance methods have the problem that they are difficult to apply to devices that perform irregular and diverse operations because they detect abnormal signs in devices that perform repetitive and regular operations.

[0006] Therefore, for devices that freely (irregularly) perform various movements, such as elevators and cranes, it is still somewhat difficult to detect abnormal signs in advance, and conventional problems caused by sudden device failures still occur.

[0007] In particular, sudden breakdowns of equipment such as elevators can lead to safety accidents that can result in casualties or cause inconvenience to users.

[0008] The present invention has been proposed to solve the above-mentioned problems, and its purpose is to provide a predictive maintenance method for a device using a cluster waveform, which collects a large amount of energy waveform information measured by the device when performing each operation from a device that freely performs various operations, classifies the collected large amount of energy waveforms by similar waveforms to form a small number of similar clusters, extracts a small number of cluster waveforms representing each similar cluster based on the similar waveforms belonging to the similar clusters, and then matches the real-time energy waveform measured from the device that freely operates in real time with the small number of cluster waveforms to detect an abnormality in the device, thereby generating an alarm and inducing management such as replacement or repair in advance before a device malfunctions, thereby preventing economic losses, safety accidents, inconveniences, etc., caused by sudden device malfunctions.

[0009] In order to achieve the above object, the predictive maintenance method of a device using a cluster waveform according to the present invention is characterized by including an information collection step (S10) of collecting information on changes in energy size over time measured by a device performing each operation in a device performing various operations, thereby collecting a large amount of energy waveform information for various operations performed by the device; a cluster waveform extraction step (S20) of classifying a large amount of energy waveforms collected in the information collection step (S10) by similar waveforms to form a plurality of similar clusters, and extracting cluster waveforms from each of the plurality of similar clusters; and a detection step (S30) of measuring and collecting real-time energy waveforms from a device performing an operation in real time, when the device performs the operation, and comparing the collected real-time energy waveform with a plurality of cluster waveform information extracted in the cluster waveform extraction step (S20) to detect abnormal signs of the device in real time.

[0010] In addition, the cluster waveform extraction step (S20) is characterized by including a classification step (S21) of classifying and collecting a large amount of energy waveforms collected in the information collection step (S10) by similar waveforms to form a plurality of similar clusters, an overlapping step (S22) of overlapping energy waveforms belonging to similar clusters in the classification step (S21) to form a cumulative waveform having a predetermined thickness for each similar cluster, a calculation step (S23) of dividing the cumulative waveform of each legacy cluster formed in the overlapping step (S22) into regular time intervals, and calculating the average energy value of the energy waveforms at the division points by summing them up, and an extraction step (S24) of extracting cluster waveforms for the cumulative waveform of each similar cluster by connecting the average values ​​of the division points calculated in the calculation step (S23).

[0011] In addition, the detection step (S30) is characterized by including a measurement process (S31) of collecting information on changes in energy size over time measured from a device performing real-time operation and measuring the energy waveform of the device for real-time operation, a matching process (S32) of searching and matching a waveform having a similar shape to the energy waveform of the device for real-time operation measured in the measurement process (S31) among a plurality of cluster waveforms, a detection process (S33) of detecting a matching rate between the real-time energy waveform of the device and the matched cluster waveform in the matching process (S32), and a judgment process (S34) of setting a threshold value for the matching rate of the detection process (S33), and determining the device as being in an unstable state and giving an alarm when the real-time matching rate between the real-time energy waveform of the device and the matched cluster waveform in the detection process (S33) is formed to be less than the threshold value.

[0012] In addition, if a waveform similar to the real-time energy waveform of the device is not searched and matched among a plurality of cluster waveforms in the matching process (S32) of the detection step (S30), the judgment process (S34) is characterized in that the real-time device status is judged to be unstable and an alarm is issued.

[0013] In addition, the energy measured and collected from the device is characterized in that any one of the following is selectively used: current consumed in operating the device, vibration generated when operating the device, noise generated when operating the device, frequency of power supplied to the device, temperature, humidity, and pressure of the device when operating the device.

[0014] As described above, according to the predictive maintenance method of a device using a cluster waveform according to the present invention, a large amount of energy waveform information measured by the device while performing each operation is collected from a device that freely performs various operations, the collected large amount of energy waveforms are classified by similar waveforms to form a small number of similar clusters, and a small number of cluster waveforms representing each similar cluster are extracted based on the similar waveforms belonging to the similar clusters, and then the real-time energy waveform measured from the device that freely operates in real time is matched and compared with the small number of cluster waveforms to detect an abnormality in the device, thereby generating an alarm and inducing management such as replacement or repair in advance before a device failure occurs, thereby preventing in advance economic losses, safety accidents, inconveniences, etc. due to sudden device failure.

[0015] FIG. 1 is a block diagram of a method for predictive maintenance of a device using a cluster waveform according to an embodiment of the present invention.

[0016] FIG. 2 to FIG. 7 are drawings for explaining a predictive maintenance method of a device using the cluster waveform shown in FIG. 1.

[0017] A method for predictive maintenance of a device using cluster waveforms according to a preferred embodiment of the present invention is described in detail with reference to the attached drawings. Detailed descriptions of known functions and configurations that may unnecessarily obscure the gist of the present invention are omitted.

[0018] FIGS. 1 to 7 illustrate a predictive maintenance method for a device using a cluster waveform according to an embodiment of the present invention. FIG. 1 is a block diagram of a predictive maintenance method for a device using a cluster waveform according to an embodiment of the present invention, and FIGS. 2 to 7 are drawings for explaining a predictive maintenance method for a device using a cluster waveform illustrated in FIG. 1.

[0019] As illustrated in Fig. 1, the predictive maintenance method (100) of a device using a cluster waveform according to an embodiment of the present invention includes an information collection step (S10), a cluster waveform extraction step (S20), and a detection step (S30).

[0020] The above information collection step (S10) is a step for collecting a large amount of energy waveform information for various operations performed by a device by collecting information on changes in energy size over time measured by the device while performing each operation.

[0021] Here, in the information collection step (S10), the energy measured and collected from the device performing various operations may be any one of the current consumed in operating the device, the vibration generated when operating the device, the noise generated when operating the device, the frequency of the power supplied to the device, the temperature, humidity, pressure, etc. of the device when operating the device. In the predictive maintenance method (100) of the device using a cluster waveform of the present invention, the current consumed in operating the device is selectively used as the energy measured from the device, but the selection and use is not limited to this current.

[0022] That is, when information on changes in the size (value) of current consumed by a device performing an operation in the above information collection step (S10) is collected over time, a current (energy) waveform such as a graph is collected. Since the device performs various operations, various types of current waveforms are collected.

[0023] Here, there are devices such as elevators and cranes that freely (irregularly) perform various movements. For convenience of explanation, in the present invention, the above device is set as an elevator that is installed in a 10-story building and can accommodate about 10 people, and in the information collection step (S10), current consumed by the elevator was measured during the elevator's various movements for 30 days to collect various types of current waveforms.

[0024] As a result of measuring and collecting current waveforms from the elevator for 30 days, approximately 8,500 current waveforms were collected.

[0025] Since the predictive maintenance method (100) of the device using the cluster waveform of the present invention detects abnormal signs of the device freely performing various operations, it is desirable to induce sufficient acquisition of current waveform information on various operations performed by the device. As described above, the current waveform was measured and collected from the elevator for a total of 30 days. However, it is of course possible to freely set the period for collecting current waveform information by considering the type of device, the surrounding environment in which the device is used, and other conditions.

[0026] Here, the various types of current waveform information collected from the device are used as a medium for extracting a small number of cluster waveforms that serve as a basis for detecting abnormal symptoms of the device in the cluster waveform extraction step (S20) to be described later. The current waveform information collected in the information collection step (S10) is assumed to have been measured and collected in an elevator operating in a normal state.

[0027] The above cluster waveform extraction step (S20) classifies a large amount of energy waveforms collected in the above information collection step (S10) by similar waveforms to form a plurality of similar clusters, and extracts cluster waveforms from each of the plurality of similar clusters. The step consists of a classification process (S21), an overlapping process (S22), a calculation process (S23), and an extraction process (S24).

[0028] The above classification process (S21) is a process of classifying and collecting a large amount of energy waveforms collected in the above information collection step (S10) by similar waveforms to form a number of similar clusters.

[0029] That is, in the information collection step (S10), a large number of current waveforms are collected and classified by similar waveforms to form a small number of clusters. As described above, about 8,500 current waveforms collected from the elevator are classified by similar waveforms, and a total of 91 similar clusters are classified, and these 91 similar clusters represent the entire current waveforms (about 8,500 current waveforms).

[0030] Figure 2 illustrates some of the few similar clusters classified by similar waveforms. When a total of 91 similar clusters are sequentially numbered as the 1st similar cluster, the 2nd similar cluster, …, the 91st similar cluster, the waveforms classified as the 20th similar cluster to the 40th similar cluster are shown.

[0031] Here, among the current waveforms of the above similar clusters, current waveforms with an abnormal shape (noise cluster) are removed, and similar clusters are formed based on meaningful current waveforms.

[0032] The above superposition process (S22) is a process of superimposing energy waveforms belonging to similar clusters in the above classification process (S21) to form a cumulative waveform having a predetermined thickness for each similar cluster.

[0033] That is, when energy waveforms belonging to similar clusters are superimposed, they are formed into a single cumulative waveform having a predetermined thickness. In the classification process (S21), about 8,500 current waveforms collected from the elevator are classified into a total of 91 similar clusters. Therefore, as illustrated in FIG. 3, in the superimposition process (S22), a total of 91 cumulative waveforms extracted from a total of 91 similar clusters will be formed.

[0034] The above-mentioned production process (S23) is a process of dividing the cumulative waveform of each heritage cluster formed in the above-mentioned overlapping process (S22) into sections at regular time intervals, and calculating the average energy value for each section point by summing and averaging the energy values ​​of the energy waveforms at the section points.

[0035] For example, as illustrated in FIG. 4, the cumulative waveform of the 32nd similar cluster is formed by overlapping seven current waveforms. The current values ​​of the seven current waveforms at each segment point are all added up and then divided by 7 to extract the average value of each segment point. If the current values ​​of each current waveform at the segment point are formed as 10, 11, 12, 13, 15, 18, and 19, the average value becomes 98 / 7=14.

[0036] Of course, the average value for each segment point is calculated for a total of 91 cumulative waveforms extracted from a total of 91 similar clusters in the above manner.

[0037] The above extraction process (S24) is a process of connecting the average values ​​of each segment point produced in the above production process (S23) to each other and extracting the cluster waveform for the cumulative waveform of each similar cluster.

[0038] That is, as shown in Fig. 5, when the average values ​​of each segment point derived from the cumulative waveform of the 32nd similar cluster are connected to each other, a cluster waveform for the cumulative waveform of the 32nd similar cluster is constructed (extracted).

[0039] Of course, since cluster waveforms are extracted for each of the cumulative waveforms of a total of 91 similar clusters, a total of 91 cluster waveforms will be formed, and the 91 cluster waveforms formed in this way will be representative of the entire current waveform.

[0040] Here, in the above-mentioned production process (S23), a reasonable method for constructing the cluster waveform from the accumulated waveform in which a plurality of current waveforms are superimposed is to extract the average value of the current values ​​of the current waveforms at each of the above-mentioned segment points, but it may also be constructed in a manner of extracting the median value of the thickness of the superimposed current waveform, the Max or Min value, the value of a selected specific location, etc.

[0041] As illustrated in FIG. 6, the detection step (S30) measures and collects real-time energy waveforms from the device performing the operation when the device performs the operation in real time, and compares the collected real-time energy waveforms with a plurality of cluster waveform information extracted in the cluster waveform extraction step (S20) to detect abnormal signs of the device in real time. The detection step (S30) consists of a measurement process (S31), a matching process (S32), a detection process (S33), and a judgment process (S34).

[0042] The above measurement process (S31) is a process of collecting information on changes in energy size over time measured from a device performing real-time operation and measuring the energy waveform of the device for real-time operation.

[0043] That is, when the elevator operates to detect the status of the elevator in real time, the current waveform measured from the elevator during the operation is measured and collected in real time.

[0044] The above matching process (S32) is a process of searching and matching a waveform having a shape similar to the energy waveform of the device for real-time operation measured in the above measurement process (S31) among a plurality of cluster waveforms.

[0045] That is, it is a process of searching and matching cluster waveforms similar to real-time current waveforms by matching real-time current waveforms measured from an elevator with a total of 91 cluster waveforms extracted in the cluster waveform extraction step (S20).

[0046] The above detection process (S33) is a process of detecting the matching rate between the cluster waveform matched with the real-time energy waveform of the device in the above matching process (S32).

[0047] That is, the real-time current waveform of the elevator and the cluster waveform matched in the above matching process (S32) are superimposed to detect and provide a matching rate between the current waveform of the elevator and the cluster waveform. Since the cluster waveform is extracted based on the current waveform collected from an elevator in a normal state, the higher the matching rate is because the current waveform of the elevator and the matched cluster waveform are similar to each other, the more the elevator's status can be judged as normal, and the lower the matching rate is, the more the elevator's status can be judged as abnormal.

[0048] The above judgment process (S34) is a process of setting a threshold value for the matching rate of the above detection process (S33), and if the real-time matching rate between the cluster waveform matched with the real-time energy waveform of the device in the above detection process (S33) is formed to be less than the threshold value, the device is judged to be in an unstable state and an alarm is issued.

[0049] That is, in the above detection process (S33), if the matching rate between the real-time current waveform of the elevator and the matched cluster waveform is detected to be greater than the set threshold value, the elevator is judged to be in a stable state.

[0050] Conversely, if the matching rate between the cluster waveform and the real-time current waveform of the elevator is detected to be below a set threshold, the elevator is judged to be in a somewhat unstable state and an alarm is generated so that management such as inspection or repair can be performed in advance before a breakdown occurs in the elevator.

[0051] Here, the threshold value set in the above judgment process (S34) can be set to various sizes of values ​​considering conditions such as the type of device, usage environment, and lifespan, and the threshold value can be divided into at least two threshold values, for example, an alarm threshold value and a risk threshold value at a level higher than the alarm threshold value, and the level of the alarm can be formed in various ways to alert for abnormal signs of the device in stages.

[0052] Of course, if the elevator freely performs various operations, the current waveform for each operation can be measured in real time from the elevator, and the status of the elevator performing each operation can be continuously detected and judged in the same manner as above, so that abnormal signs of irregularly operating equipment can be easily detected in real time and effectively preventive maintenance can be performed.

[0053] Meanwhile, as illustrated in FIG. 7, in the matching process (S32) of the detection step (S30), if a waveform similar to the real-time energy waveform of the device is not searched and matched among a plurality of cluster waveforms, the state of the real-time device is judged to be unstable and an alarm is issued in the judgment process (S34).

[0054] In other words, if a waveform similar to the real-time current waveform of the elevator does not exist in a small number of cluster waveforms (a total of 91 cluster waveforms), it means that the real-time current waveform of the elevator is a waveform of a different form from the current waveform measured in a normal elevator, and such a current waveform is sufficient reason to suspect the status of the elevator, so an alarm is issued to judge the elevator to be in a somewhat unstable state so that management such as inspection or repair can be performed in advance before a breakdown occurs in the elevator.

[0055] The predictive maintenance method (100) of the present invention using a cluster waveform, which is comprised of the above-described process, collects a large amount of energy waveform information measured from a device performing each operation from a device that freely performs various operations, classifies the collected large amount of energy waveforms by similar waveforms to form a small number of similar clusters, extracts a small number of cluster waveforms representing each similar cluster based on the similar waveforms belonging to the similar clusters, and then matches and compares the real-time energy waveform measured from the device that freely operates in real time with the small number of cluster waveforms to detect an abnormality in the device, thereby generating an alarm and inducing management such as replacement or repair in advance before a device malfunctions, thereby preventing economic losses, safety accidents, inconveniences, etc., caused by sudden device malfunctions in advance.

[0056] While the present invention has been described with reference to the embodiments illustrated in the accompanying drawings, these are merely illustrative and are not limited to the above-described embodiments. Those skilled in the art will readily appreciate that various modifications and equivalent embodiments are possible. Furthermore, it should be understood that modifications are possible by those skilled in the art without departing from the spirit and scope of the present invention. Therefore, the scope of the claims in the present invention is not limited to the detailed description, but rather by the claims set forth below and their technical spirit.

[0057] The present invention can be used in the predictive maintenance industry of equipment.

Claims

1. In a predictive maintenance method for a device that performs various movements irregularly, An information collection step (S10) that collects information on changes in energy size over time measured by the device and collects a large amount of energy waveform information on various operations performed by the device; A cluster waveform extraction step (S20) that classifies a large amount of energy waveforms collected in the above information collection step (S10) into similar waveforms to form a plurality of similar clusters, and extracts cluster waveforms from each of the plurality of similar clusters; and A predictive maintenance method for a device using a cluster waveform, characterized in that it comprises a detection step (S30) for measuring and collecting real-time energy waveforms from a device performing an operation in real time, and detecting abnormal signs of the device in real time by comparing the collected real-time energy waveforms with a plurality of cluster waveform information extracted in the cluster waveform extraction step (S20).

2. In paragraph 1, The above cluster waveform extraction step (S20) is A classification process (S21) that collects a large amount of energy waveforms collected in the above information collection step (S10) and classifies them into similar waveforms to form a number of similar clusters, In the above classification process (S21), an overlapping process (S22) is performed to overlap energy waveforms belonging to similar clusters to form a cumulative waveform having a predetermined thickness for each similar cluster, A calculation process (S23) for calculating the average energy value for each partition point by dividing the cumulative waveform of each heritage cluster formed in the above overlapping process (S22) at regular time intervals and calculating the average energy value of the energy waveforms at the partition points by adding up the average energy value, A predictive maintenance method for a device using a cluster waveform, characterized in that it includes an extraction process (S24) for extracting a cluster waveform for each cumulative waveform of each similar cluster by connecting the average values ​​of each section point calculated in the above calculation process (S23).

3. In paragraph 2, The above detection step (S30) is A measurement process (S31) for measuring the energy waveform of a device for real-time operation by collecting information on changes in energy size over time measured from a device performing real-time operation, and A matching process (S32) for searching and matching a waveform having a shape similar to the energy waveform of the device for real-time operation measured in the above measurement process (S31) among a plurality of cluster waveforms, In the above matching process (S32), a detection process (S33) for detecting the matching rate between the cluster waveform matched with the real-time energy waveform of the device, A predictive maintenance method for a device using a cluster waveform, characterized in that it includes a judgment process (S34) for determining that the device is in an unstable state and providing an alarm when a real-time matching rate between the cluster waveform matched with the real-time energy waveform of the device in the detection process (S33) is formed to be less than the threshold value for the matching rate of the detection process (S33).

4. In paragraph 3, A predictive maintenance method for a device using a cluster waveform, characterized in that if a waveform similar to the real-time energy waveform of the device is not searched and matched among a plurality of cluster waveforms in the matching process (S32) of the above detection step (S30), an alarm is issued in the judgment process (S34) judging that the real-time device status is unstable.

5. In paragraph 1, A predictive maintenance method for a device using a cluster waveform, characterized in that the energy measured and collected from the device is selected from among the current consumed in operating the device, vibration generated when operating the device, noise generated when operating the device, frequency of power supplied to the device, temperature, humidity, and pressure of the device when operating the device.

Images