Disclosure of Invention
The invention aims to provide a self-adaptive threshold value method based on extreme difference, which is characterized in that a confidence interval of a health state is continuously updated according to the existing health value of online data, when the health value exceeds the confidence interval, an abnormal value diagnosis mechanism is triggered, the possibility of the abnormal value is eliminated, and finally an optimal threshold value line is selected. The method overcomes the data dependency of the original threshold method, and generates a new confidence interval for online data every time a new sample is added, thereby enhancing the self-adaptive characteristic; in addition, the newly improved abnormal value diagnosis mechanism effectively overcomes early warning errors caused by abnormal values and enhances the accuracy of threshold early warning.
The invention is realized by adopting the following method: an adaptive thresholding method based on range, comprising:
step one, extracting an original vibration signal sample and solving a characteristic value matrix for the original vibration signal sample;
step two, solving a health value by applying a health evaluation method to the obtained characteristic value matrix;
step three, judging whether a new sample is added or not, if so, updating a confidence interval, wherein the confidence interval specifically comprises the following steps:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%;
Step four, judging whether the added new sample is in the confidence interval, wherein the method for judging whether the added new sample is in the confidence interval specifically comprises the following steps:
wherein x is
kIs the health value of the kth sample,
is the mean of the health values of the first k-1 samples; judging whether D is smaller than D, if so, judging that the added new samples are distributed in the confidence interval, otherwise, distributing the added new samples outside the confidence interval; meanwhile, storing the subscript of the new sample outside the confidence interval into a vector N, wherein the counter p is p + 1;
step five, when p is equal to n, triggering a judging mechanism, wherein n represents a preset judging number and is any integer from 1 to 10 in natural numbers; the discrimination mechanism is specifically as follows: calling a continuous number function to find a continuous number in N; if the returned continuous number is N, the returned threshold line is the health value corresponding to the N (1) element, wherein N (1) is the corresponding first sample index outside the confidence interval; if the returned continuous numbers are smaller than N, the samples corresponding to N are abnormal values, updating N, p, and repeating the third step and the fourth step to obtain an equipment degradation alarm threshold line;
step six: and expanding the range of the confidence interval, and updating the confidence interval as follows:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%;
Step seven: and repeating the fourth step and the fifth step to obtain an equipment degradation alarm threshold line.
Further, the health assessment method comprises mahalanobis distance, logistic regression, SOM-MQE method.
Further, the application method of the continuous number function is specifically as follows: the input data is a subscript vector N, whether the numerical value difference of two adjacent elements is 1 is compared by traversing each element in the N, after the traversal is finished, the continuous natural number in the N is returned, and if the continuous natural number state does not exist, the last element in the N is returned.
The invention can be realized by adopting the following system: a range-based adaptive threshold system, comprising: the device comprises an extraction module, a solving module, a judgment module I, a judgment module II, a trigger judgment module and an expansion module; the extraction module is used for extracting an original vibration signal sample and solving a characteristic value matrix for the original vibration signal sample; the solving module is used for solving a health value by applying a health evaluation method to the obtained characteristic value matrix; the first judging module is used for judging whether a new sample is added or not, and if so, updating a confidence interval, wherein the confidence interval specifically comprises:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]20%; the second judging module is configured to judge whether the added new sample is within a confidence interval, where the method for judging whether the added new sample is within the confidence interval specifically includes:
wherein x is
kIs the health value of the kth sample,
is the mean of the health values of the first k-1 samples; judging whether D is smaller than D or not,if so, judging that the added new samples are distributed in the confidence interval, otherwise, distributing the added new samples outside the confidence interval; meanwhile, storing the subscript of the new sample outside the confidence interval into a vector N, wherein the counter p is p + 1;
the trigger judging module is used for judging whether p is n or not, and triggering a judging mechanism, wherein n represents a preset judging number and is any integer from 1 to 10 in natural numbers; the discrimination mechanism is specifically as follows: calling a continuous number function to find a continuous number in N; if the returned continuous number is N, the returned threshold line is the health value corresponding to the N (1) element, wherein N (1) is the corresponding first sample index outside the confidence interval; if the returned continuous numbers are less than N, the corresponding samples in N are abnormal values, and updating N, p; the expanding module is used for expanding the range of the confidence interval, and the confidence interval is updated as follows:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%。
Further, the health assessment method comprises mahalanobis distance, logistic regression, SOM-MQE method.
Further, the application method of the continuous number function is specifically as follows: the input data is a subscript vector N, whether the numerical value difference of two adjacent elements is 1 is compared by traversing each element in the N, after the traversal is finished, the continuous natural number in the N is returned, and if the continuous natural number state does not exist, the last element in the N is returned.
Further, the characteristic values are a root mean square value, a variance, a standard deviation, a maximum value, a minimum value, an average amplitude, a kurtosis factor, a form factor, a peak value factor, a pulse index, a root mean square value, a margin factor and a skewness, respectively.
Further, the characteristic values selected by using the Fisher criterion are root mean square, variance, maximum value and minimum value.
In conclusion, the self-adaptive threshold system based on the range is provided, after vibration signals of the ball screw pair are extracted, the characteristic value matrix of the vibration signals is obtained through a series of operations, the health value of the sample is obtained by using the existing health assessment method, the confidence interval is updated once every time a new health value is obtained, and the self-adaptive characteristic of the threshold setting method is greatly improved; and then judging whether the new sample is in the confidence interval or not, and triggering a diagnosis mechanism for the sample which is not in the confidence interval to eliminate the influence of an abnormal value.
The beneficial effects are that: the self-adaptive threshold value method based on the range can be effectively applied to the health assessment of the ball screw pair, and can play a role in alarming in advance for the change of the health state of the ball screw pair to remind a user of timely processing. The threshold method updates the confidence interval in real time from the initial stage of health assessment, and the threshold judgment is carried out by using the new interval every time a new sample is added. In addition, the diagnosis of the abnormal value is newly added, so that the error of early warning caused by the abnormal value is overcome, and the accuracy of the algorithm is improved.
Detailed Description
In order to make the technical solutions in the embodiments of the present invention better understood and make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention provides an embodiment of an adaptive threshold method and system based on range, and the technical solutions in the present invention are further described in detail below with reference to the accompanying drawings:
the present invention first provides an embodiment of an adaptive threshold method based on range, as shown in fig. 1, including:
step one, S101, extracting an original vibration signal sample and solving a characteristic value matrix for the original vibration signal sample;
step two S102, a health value is solved by applying a health evaluation method to the obtained characteristic value matrix;
step three, S103, judging whether a new sample is added or not, if so, updating a confidence interval, wherein the confidence interval specifically comprises the following steps:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%;
The practical significance of the confidence interval constructed here is that for a healthy value containing only random errors, 95% of the probability is distributed in the M interval, and the probability of exceeding this range is only 5%, so that the exceeding part is considered as abnormal healthy value and starts to degrade.
Step four, S104, judging whether the added new sample is in the confidence interval, wherein the method for judging whether the added new sample is in the confidence interval specifically comprises the following steps:
wherein x is
kIs the health value of the kth sample,
is the mean of the health values of the first k-1 samples; judging whether D is smaller than D, if so, judging that the added new samples are distributed in the confidence interval, otherwise, distributing the added new samples outside the confidence interval; meanwhile, storing the subscript of the new sample outside the confidence interval into a vector N, wherein the counter p is p + 1;
step five, step 105, when p is equal to n, triggering a judging mechanism, wherein n represents a preset judging number and is any integer from 1 to 10 in natural numbers; the discrimination mechanism is specifically as follows: calling a continuous number function to find a continuous number in N; if the returned continuous number is N, the returned threshold line is the health value corresponding to the N (1) element, wherein N (1) is the corresponding first sample index outside the confidence interval; if the returned continuous numbers are smaller than N, the samples corresponding to N are abnormal values, updating N, p, and repeating the third step and the fourth step to obtain an equipment degradation alarm threshold line;
step six S106: and expanding the range of the confidence interval, and updating the confidence interval as follows:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%;
The practical significance of the confidence interval constructed here is that for a healthy value containing only random errors, there is a near 100% probability distribution in the M interval, with a probability of exceeding this range near 0%, so that the exceeding part is considered to be a complete degradation of the healthy value, beginning to fail.
Step seven S107: and repeating the fourth step and the fifth step to obtain an equipment degradation alarm threshold line.
Preferably, the health assessment method comprises mahalanobis distance, logistic regression, SOM-MQE method.
Preferably, the application method of the continuous number function is specifically as follows: the input data is a subscript vector N, whether the numerical value difference of two adjacent elements is 1 is compared by traversing each element in the N, after the traversal is finished, the continuous natural number in the N is returned, and if the continuous natural number state does not exist, the last element in the N is returned.
The input data is a subscript vector N, each element in the N is traversed to compare whether the numerical value difference of two adjacent elements is 1, after traversal is finished, the continuous number (for example, 2, 3 or 7, 8, 9 and the like) in the N is returned, if the state of the continuous number does not exist, the last element in the N is returned, the obtained threshold line is an equipment degradation alarm threshold line, and the purpose is to remind a user of the fact that the health state of the user equipment begins to change and a degradation trend already occurs.
The present invention further provides an embodiment of an adaptive threshold system based on range, as shown in fig. 2, including:
the system comprises an
extraction module 201, a
solving module 202, a
first judgment module 203, a
second judgment module 204, a
trigger judgment module 205 and an
expansion module 206; the
extraction module 201 is used for extracting an original vibration signal sample and solving a characteristic value matrix for the original vibration signal sample; the
solving module 202 is configured to apply a health evaluation method to the obtained eigenvalue matrix to solve a health value; the
first judging module 203 is configured to judge whether a new sample is added, and if so, update a confidence interval, where the confidence interval specifically is:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]20%; the
second judging module 204 is configured to judge whether the added new sample is within a confidence interval, where the method for judging whether the added new sample is within the confidence interval specifically includes:
wherein x is
kIs the health value of the kth sample,
is the mean of the health values of the first k-1 samples; judging whether D is smaller than D, if so, judging that the added new samples are distributed in the confidence interval, otherwise, distributing the added new samples outside the confidence interval; meanwhile, storing the subscript of the new sample outside the confidence interval into a vector N, wherein the counter p is p + 1;
the
trigger judging module 205 is configured to trigger a judging mechanism when p is equal to n, where n represents a preset judging number and is any integer from 1 to 10 in natural numbers; the discrimination mechanism is specifically as follows: calling a continuous number function to find a continuous number in N; if the returned continuous number is N, the returned threshold line is the health value corresponding to the N (1) element, wherein N (1) is the corresponding first sample index outside the confidence interval; if the returned continuous numbers are less than N, the corresponding samples in N are abnormal values, and updating N, p; the
expansion module 206 is configured to expand the confidence interval range, and the confidence interval is updated as:
wherein
Mean value of the health values of the first k-1 samples, d ═ max (X)
k-1)-min(X
k-1)]·20%。
Preferably, the health assessment method comprises mahalanobis distance, logistic regression, SOM-MQE method.
Preferably, the application method of the continuous number function is specifically as follows: the input data is a subscript vector N, whether the numerical value difference of two adjacent elements is 1 is compared by traversing each element in the N, after the traversal is finished, the continuous natural number in the N is returned, and if the continuous natural number state does not exist, the last element in the N is returned.
Wherein the method of the equipment degradation alarm threshold is substantially the same for different methods of health assessment, but the logistic regression is slightly different in the method of setting the equipment failure alarm threshold, as shown in fig. 3. According to the principle of logistic regression, the health values calculated by the algorithm are approximately distributed between 0 and 1, and the closer to 1, the better the health state is; the closer to 0, the worse the state of health. Therefore, the alarm threshold value for equipment failure can be artificially set between 0.1 and 0.2 according to the algorithm principle; the mahalanobis distance and the SOM-MQE are set through confidence intervals, and because the health values obtained by the two methods have large difference in magnitude in the early and late stages, the visual effect of the degradation threshold line is poor, and fig. 4-6 shows the implementation process of early degradation, such as the implementation process of mahalanobis distance shown in fig. 4 and 5, and the implementation process of SOM-MQE shown in fig. 6.
Preferably, the characteristic values are a root mean square value, a variance, a standard deviation, a maximum value, a minimum value, an average amplitude, a kurtosis factor, a form factor, a peak value, a peak factor, a pulse index, a root mean square value, a margin factor, and a skewness, respectively.
Preferably, the characteristic values selected by using the Fisher criterion are root mean square, variance, maximum value and minimum value.
In the concrete implementation of the three methods, the influence of independent variables on the method is strictly controlled according to a control variable method, the adopted data is vibration signals of a set of ball screw pairs with the same whole life cycle, the obtained vibration signals are subjected to the same pretreatment, and the total 14 characteristic values of the signals are extracted, namely a root mean square value, a variance, a standard deviation, a maximum value, a minimum value, an average amplitude value, a kurtosis factor, a waveform coefficient, a peak value factor, a pulse index, a square root amplitude value, a margin coefficient and skewness. The characteristic values selected by using the Fisher criterion are root mean square, variance, maximum value and minimum value.
In summary, the method provided by the invention is based on the range theory, the self-adaptive threshold is used for carrying out failure early warning, different health assessment methods of the ball screw pair, such as Mahalanobis distance, logistic regression and SOM-MQE, and factors such as the self-adaptive characteristic of the threshold and the irrelevant influence of an abnormal value are comprehensively considered, and the problem that large-scale historical data is relied on in the traditional problem is solved.
The above examples are intended to illustrate but not to limit the technical solutions of the present invention. Any modification or partial replacement without departing from the spirit and scope of the present invention should be covered in the claims of the present invention.