JP4371119B2 - Abnormality monitoring method - Google Patents

Description

  The present invention relates to an abnormality monitoring method for detecting the presence / absence of an abnormality of a monitoring object based on a feature amount of a waveform of an electric signal reflecting the state of the monitoring object.

  Conventionally, equipment that has a drive source such as a motor or engine among factory equipment and household equipment is to be monitored, and detects whether there is an abnormality in the monitored object based on vibrations and sound waves generated from the monitored object Are provided in various ways (for example, see Patent Document 1).

  In this type of abnormality monitoring technology, vibration and sound waves (vibration and sound waves are detected by different sensors, but the processing process is the same, so only the sound waves will be described below. Unless otherwise specified, the term “at least one of sound wave and vibration” is used to convert the signal into an electric signal using a microphone or vibration pickup, and the monitoring object operates normally or abnormally based on the characteristic amount of the waveform of the electric signal. Judgment.

The timing for starting the extraction of the feature value for the electrical signal corresponding to the sound wave that is continuously input is determined by a trigger signal generated according to the operation of the monitoring target. For example, if the monitored object is an apparatus including a rotating part such as a cutting machine, a trigger signal is generated for each rotation of the rotating part from the monitored object, and a waveform for a certain period of time from the generation of the trigger signal, By extracting the feature amount, it is possible to compare the waveform of the electric signal under the same condition every rotation, and it is possible to determine whether there is an abnormality in the operation of the monitored object by comparing the waveform. .
JP 2003-232674 A

  However, the timing at which the trigger signal is generated from the monitored object is not necessarily the timing suitable for determining whether there is an abnormality, and the waveform used to determine whether there is an abnormality at the timing of the trigger signal generated by the monitored object. It may not be obtained.

  For example, when the target waveform to be used for determining whether there is an abnormality is earlier than the timing at which the trigger signal is generated, there is a problem that the entire target waveform cannot be acquired. Conversely, when the start of the target waveform is later than the timing at which the trigger signal is generated, there may be a silent period (no signal period) in which no sound wave is generated before or after the target waveform.

  As described above, there are cases where the target waveform cannot be accurately extracted at the timing of the trigger signal generated by the monitoring target, and the accuracy of determining whether there is an abnormality in the operation may be reduced.

  Even if the waveform is in a normal state, the target waveform may expand or contract in the time axis direction depending on the operating speed of the monitored object, or an abnormal value may occur due to the influence of noise components. In some cases, it may be difficult to distinguish the waveform from the abnormal operation.

  The present invention has been made in view of the above reasons, and its purpose is to determine only a target waveform by determining a period for extracting a feature amount based on a waveform of an electrical signal reflecting the operation of a monitoring target. An object of the present invention is to provide an anomaly monitoring method that can be reliably extracted and can accurately detect whether the operation of a monitoring object is normal or abnormal.

According to the first aspect of the present invention, an electrical signal reflecting the state of the monitored object is converted into a digital signal and stored in a memory, and a feature amount reflecting the waveform characteristic of the digital signal stored in the memory is extracted and extracted. An abnormality monitoring method for detecting the presence / absence of an abnormality of a monitoring object using a feature amount, wherein a start time and an end time of a target waveform for extracting a feature amount using a time change of the level of the digital signal are obtained. determined, by using the representative value for each time interval while dividing each time that defines the target waveform segment to extract a feature amount of the waveform of the electrical signal, per to determine the presence or absence of the monitored object abnormal The digital signal is stored in the memory for a monitoring period of a certain time including the target waveform, and each determination period is divided into a determination period for each fixed time width The effective value of and is adjacent to the previous and subsequent determination periods in the time series, and the effective value of the subsequent determination period is greater than the effective value of the previous determination period, and the difference between the effective values The time between the pair having the maximum absolute value is used as the start time, and the effective value of the later determination period is greater than the effective value of the previous determination period in the adjacent determination period groups in the time series. The time between the pairs that are smaller and the absolute value of the difference between the two effective values is maximized is used as the end time .

  According to this method, since the start time and end time of the target waveform are determined using the time change of the level of the digital signal stored in the memory, the target waveform from which the feature amount is extracted can be extracted with high reproducibility. . In other words, the region representing the characteristics of the operation of the monitoring object in the electrical signal cannot be captured, or many unnecessary regions are not captured. Moreover, since the target waveform is cut out under the same conditions, it is possible to detect the presence / absence of abnormality of the monitoring target object with high accuracy.

  When the monitoring target has a function for generating a trigger signal, the trigger signal generation timing is generated sufficiently ahead of the target waveform, and the trigger signal is generated and stored in the memory. Thus, it is possible to prevent the memory storage capacity from becoming larger than necessary. In addition, when extracting feature values, the target waveform is divided into time intervals, and representative values for each time interval are used. Therefore, the amount of information is suppressed by suppressing minute changes unnecessary for extracting feature values. And the processing load when extracting the feature amount can be reduced. That is, it is possible to remove unnecessary information from the target waveform and to make the difference between the characteristic portions remarkable. As a result, the possibility that the monitoring target is in a normal state but erroneously determined as abnormal is reduced, and the abnormal state can be accurately detected.

In addition, the start time and end time of the target waveform are determined by dividing the digital signal stored in the memory for each determination period and using the time change of the effective value of each determination period. The target waveform can be extracted with high accuracy, and as a result, it is possible to accurately detect whether there is an abnormality in the monitoring target. In addition, the target waveform can be extracted even if the monitoring target does not generate a trigger signal, and even when the trigger signal is generated, the trigger signal generation timing and the target waveform are generated each time the trigger signal is generated. Even when the generation timing is different, the presence / absence of abnormality can be accurately determined by extracting the target waveform.

According to a second aspect of the present invention, an electrical signal reflecting the state of the monitored object is converted into a digital signal and stored in a memory, and a feature quantity reflecting the waveform characteristics of the digital signal stored in the memory is extracted and extracted. An abnormality monitoring method for detecting the presence / absence of an abnormality of a monitoring object using a feature amount, wherein a start time and an end time of a target waveform for extracting a feature amount using a time change of the level of the digital signal are obtained. In determining whether there is an abnormality in the monitoring target by dividing the target waveform into specified time intervals and extracting the characteristic amount of the waveform of the electrical signal using the representative value for each time interval. , storing the digital signal for monitoring period of a predetermined time including the target waveform to the memory, the decision period while dividing the digital signals stored in the memory in the determination period at predetermined time width The effective value of the subsequent determination period is larger than the effective value of the previous determination period and includes the previous determination period in the set of adjacent determination periods adjacent to each other in time series. A set satisfying the condition that the effective value in at least one of the first predetermined determination periods before the determination period is equal to or less than the first threshold value, and the absolute value of the difference between the two effective values is maximized The time between the pairs is used as the start time, and the effective value of the subsequent determination period is smaller than the effective value of the previous determination period in the determination period adjacent groups in the time series, and A set satisfying the condition that the effective value in at least one of the second predetermined number of determination periods after the determination period including the determination period is equal to or less than the second threshold value, and the difference between the effective values The time between the pair having the maximum absolute value of is used as the end time. And butterflies.

According to this method, in addition to the same operation as that of the first aspect of the invention, the conditions used for determining the start time and the end time are stricter, so that the start time and the end time can be detected more accurately. . In particular, since the effective value is compared with the first threshold value and the second threshold value to determine whether there is a no-signal period before and after the target waveform, the target waveform is extracted by removing the no-signal period. In addition, it is possible to prevent the target waveform from being erroneously cut when the target waveform includes a portion where the effective value largely changes.

According to a third aspect of the present invention, an electrical signal reflecting the state of an object to be monitored is converted into a digital signal and stored in a memory, and a feature amount reflecting the waveform characteristic of the digital signal stored in the memory is extracted and extracted. An abnormality monitoring method for detecting the presence / absence of an abnormality of a monitoring object using a feature amount, wherein a start time and an end time of a target waveform for extracting a feature amount using a time change of the level of the digital signal are obtained. In determining whether there is an abnormality in the monitoring target by dividing the target waveform into specified time intervals and extracting the characteristic amount of the waveform of the electrical signal using the representative value for each time interval. , storing the digital signal for monitoring period of a predetermined time including the target waveform to the memory, the decision period while dividing the digital signals stored in the memory in the determination period at predetermined time width After the effective value of the later determination period is larger than the effective value of the previous determination period in the set of adjacent determination periods adjacent to each other in the time series, and includes the later determination period Between the pairs that satisfy the condition that the effective value is equal to or greater than the first threshold in all of the first specified number of determination periods after the determination period, and in which the absolute value of the difference between the two effective values is maximized Is used as the start time, and the effective value of the subsequent determination period is smaller than the effective value of the previous determination period in the set of determination periods adjacent to each other in time series, and the previous determination period is A set that satisfies the condition that the effective value is greater than or equal to the second threshold value in all of the second specified number of determination periods before the previous determination period, and that maximizes the absolute value of the difference between the two effective values. Is used as the end time.

According to this method, in addition to the same operation as that of the first aspect of the invention, the conditions used for determining the start time and the end time are stricter, so that the start time and the end time can be detected more accurately. . In particular, by comparing the effective value with the first threshold value and the second threshold value, it has been confirmed that the target waveform is a period with a signal, thus reducing the possibility of accidental noise being the target signal. can do.

According to a fourth aspect of the present invention, an electrical signal reflecting the state of an object to be monitored is converted into a digital signal and stored in a memory, and a feature quantity reflecting the waveform characteristics of the digital signal stored in the memory is extracted and extracted. An abnormality monitoring method for detecting the presence / absence of an abnormality of a monitoring object using a feature amount, wherein a start time and an end time of a target waveform for extracting a feature amount using a time change of the level of the digital signal are obtained. In determining whether there is an abnormality in the monitoring target by dividing the target waveform into specified time intervals and extracting the characteristic amount of the waveform of the electrical signal using the representative value for each time interval. , storing the digital signal for monitoring period of a predetermined time including the target waveform to the memory, the decision period while dividing the digital signals stored in the memory in the determination period at predetermined time width The effective value of the subsequent determination period is larger than the effective value of the previous determination period and includes the previous determination period in the set of adjacent determination periods adjacent to each other in time series. The effective value in at least one of the first predetermined number of determination periods before the first determination period is equal to or less than the first threshold, and the second predetermined number of times after the subsequent determination period including the subsequent determination period A time series between a pair that satisfies the condition that the effective value is greater than or equal to the second threshold value in all the determination periods and that has the maximum absolute value of the difference between the two effective values is used as the start time. The effective value of the later determination period is smaller than the effective value of the previous determination period in the set of determination periods adjacent to each other in front and rear, and the third specified number of items after the subsequent determination period including the subsequent determination period The effective value in at least one of the determination periods is less than or equal to the third threshold value And a set that satisfies the condition that the effective value is greater than or equal to the fourth threshold in all of the fourth specified number of determination periods before the previous determination period including the previous determination period, and the difference between the two effective values The time between the pair having the maximum absolute value is used as the end time.

According to this method, in addition to the same operation as that of the first aspect of the invention, the conditions used for determining the start time and the end time are stricter, so that the start time and the end time can be detected more accurately. . In particular, in this method, since the target waveform is extracted when all of the conditions of claims 2 and 3 are satisfied, the target waveform is removed even when the no-signal period is removed and the change in the effective value of the target waveform is large. Can be extracted without cutting, and the possibility of accidental noise being mistaken for the target signal can be reduced.

  According to the method of the present invention, since the start time and end time of the target waveform are determined using the time change of the level of the digital signal stored in the memory, the target waveform from which the feature amount is extracted can be extracted with high reproducibility. There is an advantage that can be. In addition, since the target waveform is divided into time intervals and feature quantities are extracted using time-series data of representative values for each time interval, the amount of information can be suppressed by suppressing minute changes unnecessary for extracting feature quantities. The effect that the processing load when extracting the feature amount can be reduced, the possibility that the normal state is erroneously determined as abnormal, and that the abnormal state can be reliably detected can be reduced. Is obtained.

  In the embodiment described below, a configuration for determining normality / abnormality of an operation of a monitoring object such as a cutting machine using a neural network is illustrated. For example, it is assumed that workpieces are transported one after another on the production line, and sound waves or vibrations are intermittently generated from the monitored object, such as when drilling each workpiece with a machine tool such as a drilling machine. . The neural network is assumed to be realized by executing an appropriate application program on a sequential processing type computer, but a dedicated neurocomputer can also be used. As the neural network, it is desirable to use an unsupervised competitive learning type neural network. In addition, it is possible to determine whether the monitoring target is normal or abnormal using fuzzy logic instead of the neural network.

  In this neural network, learning is performed in advance by inputting a set of learning data, and given arbitrary input data in a learned state reflects the degree of similarity or difference between the given input data and the learning data. Output is obtained. In general, a clustering map associated with learning data is set in the output layer of the neural network, the region associated with the category of learning data in the clustering map, and the output corresponding to the given input data (fired neurons) The degree of difference between the input data and the learning data is determined according to the position occupied by in the clustering map.

  The configuration shown in FIG. 1 includes a signal input unit 1 that detects sound waves and vibrations generated by a monitoring target. The signal input unit 1 is provided with a vibration sensor 1a for detecting vibration and a microphone 1b for detecting sound waves. Only one of the vibration sensor 1a and the microphone 1b may be used according to the type of the monitoring object, or both the vibration sensor 1a and the microphone 1b may be used in combination. Here, the start and end timings of detection of sound waves and vibrations in the signal input unit 1 are determined by an external signal (for example, a trigger signal) to the input / output unit 5 described later. In the present embodiment, the vibration sensor 1a and the microphone 1b are used in the signal input unit 1, but other physical quantities or chemical quantities that reflect the operation of the monitoring target are detected using other physical sensors or chemical sensors. May be.

  The output of the signal input unit 1 is input to the A / D converter 2 and converted into a digital signal, and then input to the CPU 3 constituting the neural network. The CPU 3 is provided with a memory 4, and the output of the A / D converter 2 is temporarily stored in the memory 4. Here, the memory 4 has an area used as a register that temporarily holds digital signals (data) for a predetermined time output from the A / D converter 2. The start of data import into the memory 4 can be defined by a trigger signal generated by the monitoring target. The digital signal stored in the memory 4 is a digital signal output from the A / D converter 2 during a fixed monitoring period including the target waveform, and the monitoring period is specified by an external signal supplied to the input / output unit 5. .

  The computer constituted by the CPU 3 and the memory 4 includes a waveform extraction unit that extracts a target waveform from which a feature amount is extracted from a digital signal supplied from the A / D converter 2 and stored in the memory 4, and a time that defines the target waveform. A section dividing unit that divides each section and obtains a representative value of each time section, and a feature extraction processing section that extracts time-series data of the representative value for each time section obtained by the section dividing section to extract a waveform feature amount And a function of a neural network that determines the degree of difference from the learning data using the feature amount extracted by the feature extraction processing unit as input data. Here, it is assumed that the neural network also includes the function of the clustering map described above. In addition, the determination result obtained by the neural network is output to the outside through the input / output unit 5 to notify normality / abnormality of the operation of the monitoring target or to be used for controlling other devices.

  The section dividing unit divides the waveform of the sound wave for each time section to the extent that information necessary for performing feature extraction from the target sound wave is preserved. For example, when it is sufficient to extract the characteristics of sound waves of fHz or less, the unit of the time interval is set to ½ f (ms). Now, assuming that an electrical signal as shown in FIG. 2 is output from the microphone 1b, after the electrical signal is converted into a digital signal and stored in the memory 4, a part of the waveform is cut out as a target waveform as will be described later. The target waveform is divided for each time interval Tm. As a signal value representing each time interval Tm, an average value or a maximum value of the signal values in each time interval Tm is used.

  Therefore, when the target waveform is input to the section dividing unit, the time series data output from the section dividing unit represents a waveform from which components having a shorter period than the period of the time section Tm are removed as shown in FIG. Become. That is, the section dividing unit functions equivalent to a low-pass filter because it removes frequency components higher than the frequency necessary for feature extraction.

  Since the average value or the maximum value in the time interval Tm is used as the representative value of the time interval Tm divided by the interval division unit, the sampling period of the A / D converter 2 needs to be half or less of the time interval Tm. is there. For example, since the sampling frequency of the general audio A / D converter 2 is 40 kHz or more, if the audio A / D converter is employed and the time interval Tm is set to 1 ms, for example, the sampling period is set to the time interval Tm. Can be made sufficiently shorter.

  The section dividing unit may smooth the sound wave waveform before dividing the sound wave waveform in units of the time section Tm. For the smoothing process, for example, a moving average may be used. As described above, if the sampling period of the A / D converter 2 is sufficiently shorter than the time interval Tm, the time change of the moving average value obtained by averaging the output values of the A / D converter 2 by a certain number is also a sound wave. It includes the characteristics of the waveform. For example, if the waveform shown in FIG. 2 is smoothed as the pre-processing for dividing the time interval Tm, the waveform shown in FIG. 3 is obtained. By performing smoothing in this way, noise (for example, sudden sound) that occurs suddenly can be removed, information unnecessary for feature extraction can be reduced, and the accuracy of the final determination result can be improved. It becomes possible.

  Similarly, the section dividing unit may divide the sound wave waveform in units of time sections Tm, determine the representative value for each time section Tm, and then perform smoothing as post-processing. Since the representative value for each time interval Tm changes as shown in FIG. 3, for example, if the waveform of FIG. 3 is smoothed, a waveform without fine irregularities is obtained as shown in FIG. As a smoothing technique, a moving average value obtained by averaging a certain number of representative values obtained for each time interval Tm can be used. However, since information with the time interval Tm as a unit is lost due to this smoothing, smoothing by post-processing is used for extracting features that appear in a longer cycle than the time interval Tm. Even if smoothing is performed by post-processing as in the case of smoothing by pre-processing, unnecessary information for feature extraction can be reduced, leading to improvement in the accuracy of the final judgment result. .

  Both the pre-processing and the post-processing may be performed. However, in order to remove abnormal values, either one may be normally performed.

  By the way, as described as the background art, the determination result may differ depending on which time interval Tm of the waveform output from the interval division unit is given to the feature extraction unit. In order to avoid this type of problem, the waveform extraction unit extracts a target waveform from which a feature amount is to be extracted from the digital signal stored in the memory 4.

  As a method of cutting out the target waveform in the waveform extraction unit, as shown in FIG. 6, the waveform stored in the memory 4 is divided in units of an appropriate determination period Tp, and the time change of the effective value in each determination period Tp is evaluated. Then, a method of cutting out the target waveform by extracting the start time and the end time of the target waveform is adopted. Although the determination period Tp varies depending on the feature quantity to be extracted, the determination period Tp may be set to, for example, 10 ms if the feature quantity is extracted for the sound wave generated by the monitoring target. The determination period Tp is set to be longer than the above-described time interval Tm because it is not intended to extract the feature quantity of the target waveform, but is sufficiently shorter than the machining time for one workpiece (for example, 100). 1 or less).

  In order to extract the target waveform from the digital signal of the monitoring period stored in the memory 4, as described above, the digital signal of the monitoring period is divided for each determination period Tp, and an effective value for each determination period Tp is obtained. That is, time-series data of effective values for each determination period Tp is generated. For each pair of determination periods Tp adjacent to each other in the time series, the following evaluation is performed to extract the start time and end time of the target waveform.

  First, the absolute value of the difference between the effective values is obtained for each set of determination periods Tp adjacent to each other in the time series. Here, at the timing when the target waveform is started, as in the determination periods (3) and (4) of the determination periods Tp represented by parenthesized symbols (1) to (9) in FIG. The effective value of the period (4) is larger than the effective value of the previous determination period (3), and the absolute value of the difference between the effective values is larger (that is, the maximum) than the other determination period Tp pairs. It is thought that there are many. Although not shown, at the timing when the target waveform ends, the effective value of the subsequent determination period becomes smaller than the effective value of the previous determination period, and the absolute value of the difference between the effective values is different from the other determination periods Tp. It is thought that it is often larger (that is, maximized) than

Therefore, based on the characteristics described above, when the pair of each pair of determination periods Tp adjacent to each other in the time series is a pair that satisfies any of the conditions (a) and (b) described below, the condition ( a) When the absolute value of the difference between the effective values is maximized in each set of determination periods Tp satisfying any one of (b), the time between the determination periods Tp (that is, the boundary) is set as the start time or end time. Judge as time. When the condition (a) is satisfied, the start time is determined. When the condition (b) is satisfied, the end time is determined. The time t0 in FIG. 6 represents the start time.
(A) A set in which the effective value of the subsequent determination period Tp is larger than the effective value of the previous determination period Tp.
(B) A set in which the effective value of the subsequent determination period Tp is smaller than the effective value of the previous determination period tp.

  By the way, depending on the monitored object, the change in the amplitude of the target waveform may be large, and it is not the start time or the end time in the set of determination periods Tp that satisfy the condition (a) or (b). There may be a set in which the absolute value of the difference between the effective values in the adjacent determination periods Tp is maximum.

In such an object to be monitored, it is not sufficient to specify the start time and the end time simply by extracting a set that maximizes the absolute value of the difference between the effective values from the sets that satisfy the conditions (a) and (b). It is. Therefore, in addition to the condition of (a) as the condition of the group for determining the start time, at least one of the following conditions (c) and (d) is used, and the condition of the condition of (b) is set as the condition of the group for determining the end time: In addition, if at least one of the following conditions (e) and (f) is used, the target waveform can be cut out more accurately.
(C) A set that satisfies a condition that an effective value in at least one of three consecutive determination periods Tp before the previous determination period Tp including the previous determination period Tp is equal to or less than a predetermined threshold value.
(D) A set that satisfies the condition that the effective value is equal to or more than a predetermined threshold in all six consecutive determination periods Tp after the subsequent determination period Tp including the subsequent determination period Tp.
(E) A set that satisfies a condition that an effective value in at least one of three consecutive determination periods Tp after the subsequent determination period Tp including the subsequent determination period Tp is equal to or less than a predetermined threshold value.
(F) A set that satisfies a condition that the effective value is equal to or more than a predetermined threshold in all six consecutive determination periods Tp before the previous determination period Tp including the previous determination period Tp.

  That is, the start time is determined by any one of (a), (a) and (c), (a) and (d), (a) and (c) and (d), and (b) and (b) And the end time is determined by any of (e), (b) and (f), (b) and (e) and (f). It seems that the target waveform can be cut out more accurately as the number of conditions increases, but the time required for judgment increases as the number of conditions increases, and depending on the characteristics of the monitored object, the result is the same regardless of the conditions. How to combine the conditions is appropriately selected according to the monitoring object.

  The conditions (c) and (e) are conditions for separating the signal period in which the target waveform exists from the no-signal period. For example, in the example of FIG. When attention is paid, since three consecutive determination periods (1) to (3) including the previous determination period (3) and before the previous determination period are considered to be no signal, these three determination periods ( If at least one of the effective values of 1) to (3) is set so that the effective value is equal to or less than the threshold and the effective value of the determination period (4) exceeds the threshold, the determination period (3) and (4) are a set of determination periods Tp that satisfy the conditions (a) and (c). Since condition (c) requires a period from no signal to a signaled period, even if there are a plurality of sets that satisfy only condition (a), condition (c) must be satisfied at the same time. Can hardly be determined, and the start time can be reliably determined by selecting the pair having the maximum absolute value of the difference between the effective values from the pair in which both conditions (a) and (c) are satisfied.

  The same applies to the end condition, and the end time can be reliably determined by selecting the pair having the maximum absolute value of the difference between the effective values from among the groups that satisfy the conditions (b) and (e). .

  On the other hand, the conditions (d) and (f) are conditions for confirming that the period of the signal is not the noise generated suddenly but the target waveform that is continuously generated. Then, when focusing on the set of the determination periods (3) and (4), the effective values of the six consecutive determination periods (4) to (9) including the later determination period (4) and the subsequent determination periods are included. If the threshold is set so that all of the values exceed the threshold and the effective value of the determination period (3) does not exceed the threshold, the determination periods (3) and (4) satisfy the conditions (a) and (d). It becomes a set of determination periods Tp. Since the condition (d) requires that the signal is changed from the no-signal state to the presence-in-signal state in a plurality of determination periods in the presence / absence period, and the presence / absence of the presence / absence signal continues, the condition (a) Even if there are a plurality of sets that satisfy only the condition, the possibility that the condition (d) is satisfied is low, and the absolute value of the difference between the effective values is the largest among the sets that satisfy both the conditions (a) and (d). If a certain group is selected, the start time can be determined reliably.

  The same applies to the end condition, and the end time can be reliably determined by selecting the pair that satisfies the conditions (b) and (f) that has the maximum absolute value of the difference between the effective values. .

  In conditions (c) and (e), the effective value and the threshold are compared for three consecutive determination periods Tp, and in the conditions (d) and (f), the effective value and the threshold are compared for six consecutive determination periods Tp. However, the number of determination periods Tp for comparing the effective value and the threshold value is not particularly limited, and the threshold value is not particularly limited as long as the threshold value is set so as to be able to distinguish between a no signal and a signal. Further, the number and threshold values of the determination period Tp in which the effective value is compared with the threshold value may be different between the determination of the start time and the determination of the end time.

  As described above, when the start time t0 and the end time are determined, the feature amount of the waveform is extracted between the start time t0 and the end time, and it is determined whether it is normal or abnormal.

It is a block diagram which shows embodiment. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Signal input part 1a Vibration sensor 1b Microphone 2 A / D converter 3 CPU
4 Memory 5 Input / output section

Claims (4)

An electrical signal that reflects the state of the monitored object is converted into a digital signal and stored in a memory, a feature value that reflects the waveform characteristics of the digital signal stored in the memory is extracted, and the extracted feature value is used. An abnormality monitoring method for detecting presence / absence of an abnormality of a monitoring target object, wherein a start time and an end time of a target waveform from which a feature amount is extracted are determined using a time change of the level of the digital signal, and the target waveform is defined In order to determine whether there is an abnormality in the monitoring target by extracting the feature amount of the waveform of the electric signal using the representative value for each time interval and dividing the time interval, the constant including the target waveform is determined. For the time monitoring period, the digital signal is stored in the memory, the digital signal stored in the memory is divided into determination periods for a certain time width, and an effective value for each determination period is obtained. Among the groups of determination periods adjacent to each other in the time series, the effective value of the subsequent determination period is larger than the effective value of the previous determination period, and the absolute value of the difference between the two effective values is maximized. The time between pairs is used as the start time, and among the groups of determination periods adjacent to each other in time series, the effective value of the subsequent determination period is smaller than the effective value of the previous determination period. An abnormality monitoring method , wherein a time between a pair having a maximum absolute value of a difference between both effective values is used as the end time . An electrical signal that reflects the state of the monitored object is converted into a digital signal and stored in a memory, a feature value that reflects the waveform characteristics of the digital signal stored in the memory is extracted, and the extracted feature value is used. An abnormality monitoring method for detecting presence / absence of an abnormality of a monitoring target object, wherein a start time and an end time of a target waveform from which a feature amount is extracted are determined using a time change of the level of the digital signal, and the target waveform is defined In order to determine whether there is an abnormality in the monitoring target by extracting the feature amount of the waveform of the electric signal using the representative value for each time interval and dividing the time interval, the constant including the target waveform is determined. For the time monitoring period, the digital signal is stored in the memory, the digital signal stored in the memory is divided into determination periods for a certain time width, and an effective value for each determination period is obtained. In a set of determination periods adjacent to each other in the time series, the effective value of the subsequent determination period is larger than the effective value of the previous determination period, and is the first before the previous determination period including the previous determination period. The time between the sets that satisfy the condition that the effective value in at least one of the predetermined number of determination periods is equal to or less than the first threshold value, and the absolute value of the difference between the two effective values is maximized. The effective value of the subsequent determination period is smaller than the effective value of the previous determination period in the set of determination periods adjacent to each other in the time series in the time series, and includes the subsequent determination period. A set satisfying the condition that the effective value in at least one of the second predetermined number of determination periods after the determination period is equal to or less than the second threshold value, and the absolute value of the difference between the two effective values is maximized. abnormality monitoring, which comprises using the time between pairs in the end time Law. An electrical signal that reflects the state of the monitored object is converted into a digital signal and stored in a memory, a feature value that reflects the waveform characteristics of the digital signal stored in the memory is extracted, and the extracted feature value is used. An abnormality monitoring method for detecting presence / absence of an abnormality of a monitoring target object, wherein a start time and an end time of a target waveform from which a feature amount is extracted are determined using a time change of the level of the digital signal, and the target waveform is defined In order to determine whether there is an abnormality in the monitoring target by extracting the feature amount of the waveform of the electric signal using the representative value for each time interval and dividing the time interval, the constant including the target waveform is determined. For the time monitoring period, the digital signal is stored in the memory, the digital signal stored in the memory is divided into determination periods for a certain time width, and an effective value for each determination period is obtained. In a set of determination periods adjacent to each other in time series, the effective value of the subsequent determination period is larger than the effective value of the previous determination period, and the first after the subsequent determination period including the subsequent determination period The time between the sets that satisfy the condition that the effective value is greater than or equal to the first threshold value in all of the predetermined number of determination periods is the start time. Used, the effective value of the subsequent determination period is smaller than the effective value of the previous determination period in the set of determination periods adjacent to each other in the time series, and before the previous determination period including the previous determination period End the time between the pair that satisfies the condition that the effective value is equal to or greater than the second threshold value in all the second specified number of determination periods and that maximizes the absolute value of the difference between the two effective values. An abnormality monitoring method characterized by being used for time of day . An electrical signal that reflects the state of the monitored object is converted into a digital signal and stored in a memory, a feature value that reflects the waveform characteristics of the digital signal stored in the memory is extracted, and the extracted feature value is used. An abnormality monitoring method for detecting presence / absence of an abnormality of a monitoring target object, wherein a start time and an end time of a target waveform from which a feature amount is extracted are determined using a time change of the level of the digital signal, and the target waveform is defined In order to determine whether there is an abnormality in the monitoring target by extracting the feature amount of the waveform of the electric signal using the representative value for each time interval and dividing the time interval, the constant including the target waveform is determined. For the time monitoring period, the digital signal is stored in the memory, the digital signal stored in the memory is divided into determination periods for a certain time width, and an effective value for each determination period is obtained. In a set of determination periods adjacent to each other in the time series, the effective value of the subsequent determination period is larger than the effective value of the previous determination period, and is the first before the previous determination period including the previous determination period. The effective value in at least one of the predetermined number of determination periods is equal to or less than the first threshold value, and the effective value in all of the second predetermined number of determination periods after the subsequent determination period including the subsequent determination period. Is a group that satisfies the condition that is greater than or equal to the second threshold value, and uses the time between the pair that maximizes the absolute value of the difference between the two effective values as the start time, and is a determination period that is adjacent in time series And the effective value of the later determination period is smaller than the effective value of the previous determination period, and at least one of the third predetermined determination periods after the subsequent determination period including the subsequent determination period. The effective value of the individual is less than or equal to the third threshold and the previous judgment period Is a set that satisfies the condition that the effective value is greater than or equal to the fourth threshold value in all the fourth predetermined number of determination periods before the previous determination period including the absolute value of the difference between the two effective values. abnormality monitoring how, which comprises using the time between pairs in the end time.

Images