TW202223563A - Machine failure detection device and method - Google Patents

Abstract

A machine failure detection device includes a memory and a processor. The memory is configured to store multiple commands and multiple candidate detection parameters detected by a machine; the processor is connected to the memory and is configured to load and execute multiple instructions to: receive multiple candidate detection parameters, and calculating multiple correlation coefficients between the multiple candidate detection parameters according to the multiple candidate detection parameters; select multiple training detection parameters from multiple candidate detection parameters according to multiple correlation coefficients; and train at least one identification model according to multiple training detection parameters, so as to use the at least one identification model to determine a failure type of multiple detection information of a tested machine. In addition, a machine failure detection method is also disclosed here.

Description

Machine fault detection device and method

The present invention relates to a machine fault detection device and method.

The production schedule of today's factories tends to be complex and highly uncertain, which will result in the need to detect whether the machine is faulty and what kind of fault occurs in the production process in order to prevent the loss of the machine or the production of products. efficiency is reduced. Since the current fault detection for the machine is often not accurate, it will not only cause the failure of the machine to be detected immediately and correctly detect what kind of fault occurs in the machine, but also it is easy to misreport the normal machine as a fault. 's machine. Therefore, it will consume considerable human resources. Based on this, how to detect machine failures in real time and reduce the probability of false alarms of machine failures is an urgent problem to be solved in the art.

The invention provides a machine fault detection device, which includes a memory and a processor. The memory is used for storing a plurality of commands and a plurality of candidate detection parameters detected by the machine; the processor is connected to the memory, and is used for loading and executing a plurality of commands to: receive a plurality of candidate detection parameters, and detect according to the plurality of candidate detection parameters The parameters calculate a plurality of correlation coefficients between the plurality of candidate detection parameters; select a plurality of training detection parameters from the plurality of candidate detection parameters according to the plurality of correlation coefficients; and train at least one identification model according to the plurality of training detection parameters to utilize at least one The identification model judges the fault types of a plurality of detection information of the machine to be tested.

The invention provides a machine fault detection method. The method includes the following steps: receiving a plurality of candidate detection parameters detected by a station, and performing curve fitting on a plurality of training detection parameters among the plurality of candidate detection parameters to generate a plurality of fitting functions corresponding to the plurality of training detection parameters and use a plurality of fitting functions to adjust a plurality of parameter values of a plurality of training detection parameters, so as to train at least one identification model according to the plurality of adjusted training and detection parameters, and use the at least one identification model to determine a plurality of the machines to be tested The failure type of the detection information.

Based on the above, the machine fault detection device provided by the present invention can combine the filtering method based on feature selection and the curve fitting method to screen and adjust the parameters detected by the machine in the past, and use these screened and adjusted parameters. The parameters are used as training data, and then the recognition model is trained by using these training data. Thereby, the machine failure detection device can receive the parameters detected by the machine in real time, and use the identification model to determine whether the machine is currently faulty and what kind of failure occurs when the failure occurs according to the parameters detected in real time. In this way, the accuracy of detecting machine failures can be greatly increased, so as to prevent the occurrence of false alarms of machine failures.

FIG. 1 is a block diagram of a machine fault detection apparatus 100 according to an embodiment of the present invention. Referring to FIG. 1 , the machine failure detection apparatus 100 may include a memory 110 and a processor 120 . The memory 110 can store a plurality of commands and a plurality of candidate detection parameters detected by a machine (not shown). In addition, the machine can be a machine of any kind of process, and there is no special restriction on the process type of the machine.

In some embodiments, the machine fault detection apparatus 100 may further include a transceiver circuit (not shown), and the transceiver circuit may be connected to the memory 110 and the processor 120 .

In a further embodiment, the transceiver circuit may receive a plurality of candidate detection parameters from a plurality of sensors (not shown) corresponding to the machine, so as to store the candidate detection parameters in the memory 110 . These sensors can periodically or aperiodically detect multiple candidate detection parameters of the tool (ie, various process-related parameters detected on the tool), and can detect these candidates at the time of these detections. Detects the respective parameter values of the parameters.

It is worth noting that the parameter values of all candidate detection parameters detected by the machine at one of the detection time points can be used as a sample, and this sample can be marked with a label in advance. Multiple labels corresponding to multiple samples of a machine can include multiple fault types, wherein each fault type can be non-fault, abnormal voltage, abnormal cooling water flow, abnormal cooling water flow, abnormal door, abnormal oil pressure and other various types of machines. type of failure.

In other embodiments, the transceiver circuit may receive a plurality of candidate detection parameters from a data server (not shown) to store these candidate detection parameters in the memory 110 , wherein the data server may store a plurality of candidate detection parameters Multiple candidate detection parameters of the machine that the sensor has detected in the past.

For example, taking the machine of the centrifugal ice water host as an example, the plurality of sensors corresponding to the machine of the centrifugal ice water host may include a temperature and humidity sensor disposed on the casing, a Sensors such as the flow sensor of the water outlet, the flow sensor arranged at the ice water outlet, and the voltage sensor arranged at the current input end. These sensors can periodically or aperiodically detect candidate detection parameters such as ambient temperature, ice water inlet temperature, ice water outlet temperature, and actual line current voltage. Thereby, the sensors can transmit the candidate detection parameters detected at multiple detection time points to the transceiver circuit or to the transceiver circuit through the data server, and the transceiver circuit stores the candidate detection parameters in the memory 110.

In some embodiments, the memory 110 is, for example, any type of random access memory (RAM), read-only memory (ROM), flash memory (flash memory), Hard disk or similar or a combination of the above.

In some embodiments, the processor 120 is, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose microprocessors (microprocessors), digital signal processors (digital signal processors) , DSP), programmable controllers, application specific integrated circuits (ASIC) or other similar devices or combinations of these devices. In this embodiment, the processor 120 can load a plurality of instructions from the memory 110 to execute the machine fault detection method of the embodiment of the present invention.

Furthermore, the processor 120 may be communicatively connected to the memory 110 . Regarding the above communication connection method, the processor 120 can be connected to the memory 110 in a wired or wireless manner, which is not particularly limited.

For the wired mode, the processor 120 may utilize universal serial bus (USB), RS232, universal asynchronous receiver/transmitter (UART), internal integrated circuit (I2C) , serial peripheral interface (serial peripheral interface, SPI), display port (display port), thunderbolt (thunderbolt) or local area network (local area network, LAN) interface for wired communication connection, and there are no special restrictions. For the wireless mode, the data processing device 120 may use a wireless fidelity (Wi-Fi) module, a radio frequency identification (RFID) module, a Bluetooth module, an infrared module, a near- There is no particular limitation on the wireless communication connection between a near-field communication (NFC) module or a device-to-device (D2D) module.

FIG. 2 is a flowchart of a method for detecting a machine failure according to some exemplary embodiments of the present invention. Referring to FIG. 1 and FIG. 2 at the same time, the method of this embodiment is applicable to the machine failure detection apparatus 100 in FIG. 1 . The following describes the implementation of the present invention with the operation relationship between the components in the machine failure detection apparatus 100 . Example of the detailed steps of the machine fault detection method.

First, in step S201, the processor 120 may receive a plurality of candidate detection parameters, and calculate a plurality of correlation coefficients between the plurality of candidate detection parameters according to the plurality of candidate detection parameters. Specifically, the processor 120 may receive a plurality of candidate detection parameters pre-stored in the memory 110 from the memory 110 . In this way, the processor 120 can calculate a plurality of correlation coefficients between these candidate detection parameters.

In some embodiments, the processor 120 may calculate a plurality of Pearson correlation coefficients between the plurality of candidate detection parameters by a filter based on feature selection method.

For example, the processor 120 may calculate the Pearson correlation coefficient between the first candidate detection parameter and the second candidate detection parameter in the plurality of candidate detection parameters based on the filtering method of feature selection (ie, using The detected multiple first parameter values of the first candidate detection parameter and multiple second parameter values of the second candidate detection parameter to calculate the Pearson correlation coefficient between the first candidate detection parameter and the second candidate detection parameter) , and calculate the Pearson correlation coefficient between the first candidate detection parameter and the third candidate detection parameter in the multiple candidate detection parameters (that is, using the multiple first candidate detection parameters detected at all detected time points A parameter value and a plurality of third parameter values of the third candidate detection parameter to calculate the Pearson correlation coefficient between the first candidate detection parameter and the third candidate detection parameter). By analogy, the processor 120 may calculate the Pearson correlation coefficients between the first candidate detection parameter and the remaining candidate parameters, and calculate the Pearson correlation coefficients between the remaining candidate detection parameters.

Next, in step S203, the processor 120 may select a plurality of training detection parameters from a plurality of candidate detection parameters according to a plurality of correlation coefficients. In other words, the processor 120 may screen the plurality of candidate detection parameters according to the plurality of correlation coefficients to select several specific candidate detection parameters, and use these specific candidate detection parameters as the plurality of training detection parameters for training.

In some embodiments, the processor 120 may select a plurality of first correlation coefficients from a plurality of correlation coefficients, and select a plurality of training detection parameters corresponding to the plurality of first correlation coefficients from a plurality of candidate detection parameters, wherein the plurality of first correlation coefficients The correlation coefficient is not greater than the correlation coefficient threshold (eg, the correlation coefficient threshold stored in the memory 110 in advance). Specifically, the processor 120 may determine which correlation coefficients among the plurality of correlation coefficients are not greater than the correlation coefficient threshold, and use these correlation coefficients not greater than the correlation coefficient threshold as the plurality of first correlation coefficients. Thereby, the processor 120 can determine which candidate detection parameters among the plurality of candidate detection parameters correspond to the plurality of first correlation coefficients, and use these candidate detection parameters corresponding to the plurality of first correlation coefficients as the plurality of training detection parameters.

Finally, in step S205 , the processor 120 can train at least one identification model according to the plurality of training detection parameters, so as to use the at least one identification model to determine the failure type of the plurality of detection information of the machine under test. Specifically, the processor 120 can train at least one identification model by at least one machine learning method according to a plurality of training detection parameters, and store the at least one identification model in the memory 110 . Therefore, when the memory 110 receives a plurality of detection information of the device under test through the transceiver, the processor 120 can read the at least one identification model and the plurality of detection information from the memory 110, and use the at least one identification model Determine the fault type of multiple detection information of the machine under test.

It should be noted that the number of machine learning methods and the number of identification models of the present invention may be one or more, and there is no particular limitation.

In some embodiments, the above-mentioned machine learning method may be a multiclass decision forest algorithm, a multiclass decision jungle algorithm, a deep neural network algorithm, and a Logis regression algorithm. Machine learning methods such as logistic regression algorithms or one vs all regression algorithms.

In some embodiments, the processor 120 may further perform curve fitting on the plurality of training detection parameters to generate a plurality of fitting functions corresponding to the plurality of training detection parameters. In this way, the processor 120 can adjust the plurality of parameter values of the plurality of training detection parameters by using the plurality of fitting functions, so as to train at least one identification model according to the plurality of adjusted training detection parameters.

Specifically, the processor 120 may calculate a fitting function corresponding to one of the training detection parameters according to all parameter values of the training detection parameter. In this way, the processor 120 can determine whether there is at least one specific parameter value among the parameter values, wherein there is at least one difference value greater than the parameter threshold value between the at least one specific parameter value and the fitting function. Therefore, the processor 120 can adjust at least one specific parameter value to the at least one parameter value corresponding to the fitting function.

By analogy, the processor 120 can calculate the fitting functions corresponding to the remaining training detection parameters in the same way, and adjust the parameter values of the remaining training detection parameters by using the fitting functions corresponding to the remaining training detection parameters. In this way, the processor 120 can generate a plurality of adjusted training detection parameters, so as to train at least one identification model according to the plurality of adjusted training detection parameters.

Further, the processor 120 may perform data processing on the plurality of samples through the above-mentioned filtering method based on feature selection and the curve fitting method to generate a plurality of training samples corresponding to the plurality of adjusted training detection parameters. Thereby, the processor 120 can train at least one recognition model by at least one machine learning method according to the training samples.

In some embodiments, the processor 120 may receive a plurality of detection parameters detected by the device to be tested, and calculate a plurality of correlation coefficients to be tested among the plurality of detection parameters according to the plurality of detection parameters. Thereby, the processor 120 can select a plurality of detection parameters to be tested from a plurality of detection parameters according to a plurality of correlation coefficients to be tested, and use at least one identification model to determine a plurality of detections of the machine to be tested according to the plurality of detection parameters to be tested The failure type of the information.

In a further embodiment, the processor 120 may perform curve fitting on a plurality of test parameters to be tested to generate a plurality of test fitting functions corresponding to the plurality of test parameters to be tested. In this way, the processor 120 can use the plurality of fitting functions to be tested to adjust the values of the parameters to be tested of the detection parameters to be tested, and use the at least one identification model to determine the parameters to be tested according to the adjusted detection parameters to be tested The fault type of the multiple detection information of the machine.

In other words, after a plurality of sensors (not shown) corresponding to the device to be tested detect the device to be tested, the processor 120 can receive a plurality of detection parameters detected by the device to be tested via the sensors. In this way, the processor 120 can also filter and adjust these detection parameters by the above-mentioned filtering method and curve fitting method based on feature selection to generate a plurality of adjusted detection parameters to be tested, which are based on the plurality of adjusted detection parameters. The test parameters to be tested are used to determine the fault type of a plurality of test information of the machine to be tested by using at least one identification model.

In some embodiments, the machine fault detection apparatus 100 may further include a display (not shown), and the display may be connected to the processor 120 and used to display the fault types of a plurality of detection information of the machine to be tested.

In some embodiments, when the number of machines is plural, the machine fault detection apparatus 100 may perform data processing on the plurality of candidate detection parameters detected by each machine based on the filtering method and the curve fitting method based on feature selection, respectively to At least one identification model is trained based on the processed candidate detection parameters.

It is worth noting that the above-mentioned machine fault detection method can also be applied to error detection of other data analysis, and is not limited to the detection parameters of the machine.

Through the above steps, the machine fault detection apparatus 100 according to the embodiment of the present invention can screen a plurality of candidate detection parameters of the machine based on the filtering method of feature selection, so as to generate a plurality of training detection parameters, and use the curve fitting method to match the detection parameters. A plurality of training detection parameters are adjusted to train an identification model according to the plurality of adjusted training detection parameters. In this way, the machine failure detection apparatus 100 according to the embodiment of the present invention can use the identification model to determine the failure types of a plurality of detection information of the machine to be tested, so as to prevent the occurrence of false alarms of machine failures.

FIG. 3 is a flowchart of a method for detecting machine failure according to other exemplary embodiments of the present invention. Referring to FIG. 1 and FIG. 3 simultaneously, in step S301 , the processor 120 may generate a plurality of training detection parameters by using a filtering method based on feature selection according to the plurality of candidate detection parameters. In other words, the processor 120 may filter a plurality of candidate detection parameters detected by the machine based on the filtering method of feature selection to generate a plurality of training detection parameters.

In some embodiments, the parameter values of all candidate detection parameters detected by the machine at one of the detection time points may be used as a sample, and a label may be pre-marked in this sample. Multiple labels corresponding to multiple samples of the machine can include multiple fault types

For example, taking the centrifugal ice water host machine as an example, when the processor 120 calculates the correlation coefficient between the ice water inlet temperature and the cooling water inlet temperature in the plurality of training detection parameters is greater than a correlation coefficient threshold , the processor 120 may delete the ice water inlet temperature and the cooling water inlet temperature among the plurality of training detection parameters. In this way, the processor 120 can filter candidate detection parameters with low correlation from the plurality of candidate detection parameters to use these selected candidate detection parameters as a plurality of training detection parameters.

Next, in step S303, the processor 120 may use a curve fitting method to adjust a plurality of training detection parameters.

For example, FIG. 4 is a schematic diagram of the voltage of the actual line current detected at multiple detected time points according to some exemplary embodiments of the present invention. Referring to Figure 4, also taking the centrifugal ice water host machine as an example, the fitting curve curve is the curve of the fitting function calculated according to the voltage of the actual line current. Based on the fitting curve curve, it can be determined that the voltage of the actual line current at 45 seconds is abnormal (that is, at 45 seconds, the difference between the voltage value of the actual line current and the voltage value on the fitting curve curve is greater than one preset parameter threshold). In other words, the voltage of the actual line current has an abnormal voltage abnormal at 45 seconds. In this way, the voltage value of the abnormal voltage can be adjusted to the voltage value on the fitting curve curve at 45 seconds.

Then, referring back to FIG. 1 and FIG. 3 at the same time, in step S305, the processor 120 may train at least one identification model according to the plurality of adjusted training detection parameters.

In some embodiments, the processor 120 may perform data processing on the plurality of samples through the above-described filtering method based on feature selection and the curve fitting method to generate a plurality of training samples corresponding to the plurality of adjusted training detection parameters. Thereby, the processor 120 can train at least one recognition model by at least one machine learning method according to the training samples.

Finally, in step S307, the processor 120 can use at least one identification model to determine the failure types of the plurality of detection information of the machine under test.

In some embodiments, after a plurality of sensors corresponding to the device to be tested detect the device to be tested, the processor 120 may receive a plurality of detection parameters detected by the device to be tested via the sensors. In this way, the processor 120 can also filter and adjust these detection parameters by the above-mentioned filtering method and curve fitting method based on feature selection to generate a plurality of adjusted detection parameters to be tested, which are based on the plurality of adjusted detection parameters. The test parameters to be tested are used to determine the fault type of a plurality of test information of the machine to be tested by using at least one identification model.

To sum up, the machine fault detection device provided by the present invention can combine the filtering method based on feature selection and the curve fitting method to screen and adjust the parameters detected by the machine in the past, and make these screened and adjusted parameters. The latter parameters are used as training data, and then the recognition model is trained by using these training data. Similarly, the machine fault detection device provided by the present invention can further combine the filtering method based on feature selection and the curve fitting method to screen and adjust the parameters to be tested detected by the machine to be tested, and then according to the screening and adjustment. The parameters to be tested are then used to determine the fault type of the machine to be tested by using the trained identification model. In this way, misjudgment of machine failures can be avoided, maintenance costs can be reduced, and the accuracy of failure types can be further improved.

Although the present invention has been disclosed above by the embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field can make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, The protection scope of the present invention shall be determined by the scope of the appended patent application.

100: Machine fault detection device 110: Memory 120: Processor S201~S205, S301~S307: the steps of the machine fault detection method curve: fitting curve abnormal: abnormal voltage

FIG. 1 is a block diagram of a machine failure detection apparatus according to an embodiment of the present invention. FIG. 2 is a flowchart of a method for detecting a machine failure according to some exemplary embodiments of the present invention. FIG. 3 is a flowchart of a method for detecting machine failure according to other exemplary embodiments of the present invention. FIG. 4 is a schematic diagram of the voltage of the actual line current detected at multiple detected time points, according to some exemplary embodiments of the present invention.

S201~S205: the steps of the machine fault detection method

Claims (10)

A machine fault detection device, comprising: a memory for storing a plurality of commands and a plurality of candidate detection parameters detected by a machine; a processor, connected to the memory, for loading and executing the instructions to: receiving the candidate detection parameters, and calculating a plurality of correlation coefficients between the candidate detection parameters according to the candidate detection parameters; selecting a plurality of training detection parameters from the candidate detection parameters according to the correlation coefficients; and At least one identification model is trained according to the training detection parameters, so as to use the at least one identification model to determine a fault type of a plurality of detection information of a machine under test. The machine failure detection device as claimed in claim 1, wherein the processor is further configured to: Select a plurality of first correlation coefficients from the correlation coefficients, and select the training detection parameters corresponding to the first correlation coefficients from the candidate detection parameters, wherein the first correlation coefficients are not greater than a correlation coefficient threshold; performing curve fitting on the training detection parameters to generate a plurality of fitting functions corresponding to the training detection parameters; and A plurality of parameter values of the training detection parameters are adjusted by the fitting functions, so as to train the at least one identification model according to the plurality of adjusted training detection parameters. The machine failure detection device as claimed in claim 1, wherein the processor is further configured to: receiving the detection parameters detected by the device to be tested, and calculating a plurality of correlation coefficients to be tested between the detection parameters according to the detection parameters; selecting a plurality of detection parameters to be tested from the detection parameters according to the correlation coefficients to be tested; and The at least one identification model is used to determine the failure type of the detection information of the machine to be tested according to the detection parameters to be tested. The apparatus for detecting machine failure as claimed in claim 3, wherein the processor is further configured to: performing curve fitting on the detection parameters to be tested to generate a plurality of fitting functions to be tested corresponding to the detection parameters to be tested; A plurality of parameter values to be measured of the detection parameters to be measured are adjusted by the fitting functions to be measured, and the at least one identification model is used to determine the detection information of the machine to be measured according to the parameter values to be measured. the fault type. The machine failure detection device as described in claim 1, further comprising: A display is connected to the processor and used to display the fault type of the detection information of the machine to be tested. A machine fault detection method, comprising: receiving a plurality of candidate detection parameters detected by a machine, and performing curve fitting on a plurality of training detection parameters among the candidate detection parameters to generate a plurality of fitting functions corresponding to the training detection parameters; and The fitting functions are used to adjust a plurality of parameter values of the training and detection parameters, so as to train the at least one identification model according to the plurality of adjusted training and detection parameters, and use the at least one identification model to determine the number of machines to be tested A fault type for each detection message. The method for detecting machine failures as described in claim 6, further comprising: before performing the curve fitting on the training detection parameters in the candidate detection parameters, calculating a plurality of correlation coefficients between the candidate detection parameters according to the candidate detection parameters; and The training detection parameters are selected from the candidate detection parameters according to the correlation coefficients. The machine fault detection method according to claim 7, wherein the step of selecting the training detection parameters from the candidate detection parameters according to the correlation coefficients comprises: A plurality of first correlation coefficients are selected from the correlation coefficients, and the training detection parameters corresponding to the first correlation coefficients are selected from the candidate detection parameters, wherein the first correlation coefficients are not greater than a correlation coefficient threshold. Curve fitting is performed on a plurality of first historical detection information corresponding to each of the parameter types in the historical detection information to generate the fitting functions corresponding to the parameter types. The machine fault detection method according to claim 6, wherein the step of using the fitting functions to adjust the parameter values of the training detection parameters to train the at least one identification model according to the parameter values includes: performing curve fitting on the training detection parameters to generate a plurality of fitting functions corresponding to the training detection parameters; and The parameter values of the training detection parameters are adjusted by the fitting functions, so as to train the at least one identification model according to the adjusted training detection parameters. The machine failure detection method as claimed in claim 6, wherein the processor is further configured to: receiving the detection parameters detected by the device to be tested, and calculating a plurality of correlation coefficients to be tested between the detection parameters according to the detection parameters; selecting a plurality of detection parameters to be tested from the detection parameters according to the correlation coefficients to be tested; performing curve fitting on the detection parameters to be tested to generate a plurality of fitting functions to be tested corresponding to the detection parameters to be tested; and Use the fitting functions to be tested to adjust the values of the parameters to be tested of the detection parameters to be tested, and use the at least one identification model to determine the values of the parameters to be tested according to the adjusted detection parameters to be tested Check the information for this failure type.

Images