WO2019035279A1

WO2019035279A1 - Artificial intelligence algorithm

Info

Publication number: WO2019035279A1
Application number: PCT/JP2018/023934
Authority: WO
Inventors: 義範宮前; 光治谷内
Original assignee: ローム株式会社
Priority date: 2017-08-18
Filing date: 2018-06-25
Publication date: 2019-02-21
Also published as: JP7012086B2; JPWO2019035279A1

Abstract

Provided is a machine health monitoring system that is easily introduced. An artificial intelligence chip 10, in which a new artificial intelligence algorithm is installed, includes: a preprocessing unit 11 which generates a feature vector by extracting a feature amount of each frequency band from input data by using a plurality of bandpass filters connected in parallel; a classifier 12 which obtains a value of a kernel function by using the feature vector and a support vector; and a post-processing unit 13 which detects an abnormality of the input data from the value of the kernel function. This artificial intelligence chip 10 functions as one element of the machine health monitoring system introduced to, for example, a smart factory by being mounted on a sensor node together with a sensor or a communication unit.

Description

Artificial intelligence algorithm

The invention disclosed herein relates to artificial intelligence algorithms.

In recent years, a machine health monitoring system has been proposed which detects an abnormality of a monitored device using a sensor node.

In addition, as an example of the prior art relevant to an artificial intelligence algorithm, a nonpatent literature 1 and a nonpatent literature 2 can be mentioned, for example.

However, in the conventional machine health monitoring system, abnormality detection of the monitoring target device is performed by collecting and analyzing measurement data of the sensor node by the server. Therefore, the amount of communication exchanged between the sensor node and the server is very large, which has been a factor that hinders the introduction of the system.

In view of the above problems found by the inventors of the present invention, the invention disclosed herein has an easy-to-install machine health monitoring system, a sensor node used therefor, an artificial intelligence chip, And, it aims at providing an artificial intelligence algorithm.

The artificial intelligence algorithm disclosed in the present specification generates a feature vector by extracting feature quantities for each frequency band from input data using a plurality of band pass filters connected in parallel; Determining a value of a kernel function using the feature vector and the support vector (first configuration).

The artificial intelligence algorithm according to the first configuration may be configured (second configuration) to further include a step of detecting abnormality of the input data from the value of the kernel function.

Further, in the artificial intelligence algorithm having the first or second configuration, the input data may be vibration data (third configuration).

Further, in the artificial intelligence algorithm having any one of the first to third configurations, it is assumed that the kernel function is a linear kernel, a Gaussian kernel, or a RBF [radial base function] kernel (fourth configuration) Good.

Also, the artificial intelligence chip disclosed in the present specification generates a feature vector by extracting feature quantities for each frequency band from input data using a plurality of band pass filters connected in parallel. A configuration (fifth configuration) includes a processing unit and a classifier for obtaining a value of a kernel function using the feature vector and the support vector.

Preferably, the artificial intelligence chip having the fifth configuration further includes a post-processing unit (a sixth configuration) for detecting an abnormality of the input data from the value of the kernel function.

Further, in the artificial intelligence chip having the fifth or sixth configuration, the classifier may be a configuration (seventh configuration) that is an OCSVM [one class support vector machine] configured by hardware.

Further, a sensor node disclosed in the present specification includes: a sensor; an artificial intelligence chip having any of the fifth to seventh configurations for receiving the input data from the sensor; the artificial intelligence chip and the server And a communication unit that performs wireless communication among the above (eighth configuration).

In the sensor node having the eighth configuration, the sensor may be a vibration sensor (ninth configuration).

The sensor node having the eighth or ninth configuration includes an environmental power generation unit, a storage unit for storing power generated by the environmental power generation unit, and each unit of the sensor node using the generated power or stored power of the storage unit. And a power management unit for supplying power to the circuit (10th configuration).

In addition, the machine health monitoring system disclosed in the present specification includes a sensor node having the above-described tenth configuration attached to a monitoring target device, and a server that receives an abnormality flag from the sensor node (a 11)).

In the machine health monitoring system according to the eleventh configuration, the server may be configured (12th configuration) to receive an abnormal flag from the sensor node and report an abnormal state.

According to the invention disclosed herein, it is possible to provide an easy-to-install machine health monitoring system, and a sensor node, an artificial intelligence chip, and an artificial intelligence algorithm used therefor.

Diagram showing the first example of equipment maintenance method (TBM) Diagram showing the second example (CBM) of equipment maintenance method Diagram showing problems when adopting Wi-SUN Figure showing an outline of an artificial intelligence chip Diagram showing an example of a monitored device Diagram showing a comparison example of anomaly detection operation by new and old artificial intelligence algorithms A diagram showing an exemplary configuration of an artificial intelligence chip A diagram showing an example of the configuration of a kernel arithmetic processing unit A diagram showing an exemplary configuration of a vector computing unit Diagram showing a configuration example of a machine health monitoring system

<Application target and system specifications>
First, the background of the artificial intelligence chip disclosed in the present specification (hereinafter referred to as AI-SNP [artificial intelligence-sensor node processor] as appropriate) will be described. The motivation for the development of AI-SNP is derived from the smart factory strategy as proposed in Industry 4.0. In such a next-generation factory, it is necessary to keep checking equipment and equipment in real time without stopping operation all day. However, on the other hand, in order to maintain the high profit of the factory, it is also necessary to consider cost reduction.

FIG. 1 and FIG. 2 are diagrams each showing an example of a facility maintenance method in a factory. Note that FIG. 1 shows an example of adopting the TBM [time-based-maintenance] method (= equivalent to the old factory). In the TBM method, repair or replacement is performed when the operation time of the device meets a predetermined standard. On the other hand, FIG. 2 shows an example of adoption of the CBM [condition-based-maintenance] method (= equivalent to a smart factory). In the CBM method, repair or replacement is performed when the state of the device meets a predetermined standard.

In the TBM method shown in FIG. 1, it is necessary to periodically maintain the equipment of the factory 200, and even when the equipment breaks down, staff members are dispatched from the headquarters 100 to the factory 200 after the fact. Therefore, the operation will be suspended until the repair or replacement of the device is completed. On the other hand, in the case of the CBM method of FIG. 2, when a sign of failure appears in the device of the factory 200, an abnormal state report from the factory 200 to the head office 100 is performed. Thus, the staff can make repairs before the device shuts down.

Thus, the CBM method (FIG. 2) is one of the best solutions that matches the smart factory strategy described above, as compared to the TBM method (FIG. 1). If the CBM method is adopted, periodic maintenance of the device is not necessary, and it is sufficient to check the device state only when an abnormal state of the device (= sign of failure) is detected.

In order to introduce a machine health monitoring system that adopts the CBM method into an existing factory, it is necessary to add a sensor node for each device, or replace the old device with a new device at a high cost. You must. There are many devices currently in operation around the world. Therefore, it is best if sensor nodes can be added to the running device without the need for additional wiring or power. However, the existing wireless modules do not necessarily have sufficient communication bandwidth as compared to the size of measurement data (raw data) obtained at the sensor node.

FIG. 3 is a view showing problems when adopting a Wi-SUN [wireless smart utility network] as a wireless module of the sensor node. In addition, Wi-SUN is one of the best solutions for realizing IIoT [industorial internet of things], and can achieve good power saving and long distance communication.

However, for example, when the vibration of the monitoring target device is detected by the sensor node and the vibration data (.about.10 kHz, 16 bits) is transmitted as raw data from the factory 200 to the main unit 100, the transmission data amount is It becomes 60 kB / s in X axis, Y axis, Z axis). This is a considerably large value compared to the communication bandwidth of Wi-SUN (100 kbps → 12.5 kB / s). As described above, when it is intended to transmit vibration data as it is raw data, the communication speed becomes too slow relative to the amount of transmission data, and the power consumption of Wi-SUN also becomes large (about 180 mW). Therefore, data compression (= reduction of the amount of transmission data) is required.

Also, in order to perform real-time sensing by a sensor node without using a battery, such as EN-OCEAN, for example, the power consumption of the wireless module must be dramatically reduced. With the AI-SNP disclosed herein, it is possible to provide a very useful machine health monitoring system deployment method that does not require additional wiring and batteries, and thus, It will be possible to dramatically accelerate the new and old devices IIoT around the world.

<AI-SNP>
In the following, AI-SNPs for machine health monitoring systems and artificial intelligence algorithms (hereinafter referred to as AI algorithms) implemented therein will be described. The applicant continues to study AI algorithms for machine health monitoring systems using open data and real data from their own factories. In addition, through this research, the applicant of the present application has obtained not only a knowledge of the AI algorithm but also a lot of knowledge of data collection know-how.

As the AI algorithm, one class support vector machine (hereinafter referred to as OCSVM [one class support vector machine]), spiking neural network, or convolutional neural network are known, and the options are not limited at all. It is not a thing. In the following, based on the results of long-term research conducted by the applicant of the present application, an example in which OCSVM is adopted as an AI algorithm candidate suitable for implementation on AI-SNP will be described. The details will be described later, and only the outline will be described here.

FIG. 4 is a diagram showing an outline of an artificial intelligence chip (= corresponding to the above-described AI-SNP) mounted on a sensor node for a machine health monitoring system. The artificial intelligence chip 10 (hereinafter referred to as an AI chip 10) of this configuration example is a semiconductor chip capable of executing an AI algorithm based on input data (such as vibration data and temperature data) from a sensor. And a pre-processing unit 11, a classifier 12, and a post-processing unit 13. The AI chip 10 is required to have low power consumption and a small area to the extent that it can be mounted on a sensor node. In addition, the above-mentioned AI algorithm is designed to realize functions such as abnormality diagnosis, toolware evaluation, or life expectancy of a monitoring target device.

A preprocessing unit (pre-processor) 11 extracts feature quantities for each frequency band from input data (for example, raw vibration data obtained by a vibration sensor) using a plurality of band pass filters connected in parallel. By doing this, a feature vector is generated. In addition, as the above-mentioned feature quantity, for example, after acquiring FFT [fast Fourier transform] amplitude of a frequency spectrum every 50 Hz from 1 Hz to 20 kHz (200 dim), respective root mean square (RMS [root mean square] ) Should be calculated.

A classifier 12 uses a feature vector input from the pre-processing unit 11 and a support vector stored in advance, and uses a kernel function (eg, linear kernel, Gaussian kernel, or RBF [radial base function] Find the kernel) value. As the classifier 12, for example, it is desirable to preferably use an OCSVM configured by hardware. Further, as the support vector, for example, input data less than 33 hours after the monitoring target device starts operation may be used for learning, and a hyperplane for OK / NG determination may be created.

A post-processor (post-processor) 13 performs abnormality detection processing of input data from the value of the kernel function obtained by the classifier 12 (= calculation of an abnormality flag indicating whether the monitored device is normal or abnormal) Process).

Thus, in the AI chip 10 of this configuration example, the AI algorithm described above is implemented using the preprocessing unit 11, the classifier 12, and the post-processing unit 13.

OCSVM used as the classifier 12 is a kind of support vector machine, and is a lightweight and practical AI algorithm by unsupervised learning. OCSVM detects one class problem on one piece of software and is not similar to the machine health monitoring system. The reason is that OCSVM itself is poor in affinity with complex and high-speed time-series data.

In recent years, studies have also been made to apply OCSVM by applying appropriate pre-processing to time-series data (see, for example, Non-Patent Documents 1 and 2). However, in any of the studies, complex preprocessing is required to extract feature quantities from time-series data, and there is room for further improvement in the implementation on sensor nodes where low power consumption and area saving are required. .

Therefore, in the present specification, an AI chip 10 is proposed in which the application of OCSVM is possible by applying a simple pre-processing as compared to the conventional method. In the pre-processing unit 11, raw input data is simply processed using only FFT calculation in order to obtain feature quantities (RMS of the frequency spectrum obtained from 1 Hz to 20 kHz every 50 Hz from 1 Hz to 20 kHz). Of course, it is difficult to perform an FFT operation with power and chip size that are appropriate for the AI chip 10. Therefore, in the preprocessing unit 11, a simple analog band pass filter is used (details will be described later).

Also, focusing on the classifier 12, it is a simple OCSVM using the kernel method. As the kernel function, as described above, it is possible to select a linear kernel, a Gaussian kernel, an RBF kernel or the like. For example, if an RBF kernel can be selected, the AI algorithm by OCSVM becomes powerful. However, in order to do so, an additional function circuit (logarithmic calculation circuit) is required, so that point needs to be noted.

If attention is focused only on analysis of the abnormal state, neglect, and learning, then only the implementation of the adder, the multiplier and the simple kernel function f (X) expressed by the following equation (1) Just do it.

Therefore, in the present specification, an example in which the classifier 12 (especially its kernel operation processing unit) is configured using an adder and a multiplier will be described. Also, as a device for implementing the above AI algorithm on the AI chip 10, many small-scale analog PEs are used to reduce AD / DA [analog-to-digital / digital-to-analog] and eliminate high-speed clock. It is desirable to arrange [processing engine] so that computing units and memories have an analog structure.

Next, an example using open data to detect an abnormality occurring in a bearing of a milling machine according to an AI algorithm proposed by the applicant will be described.

FIG. 5 is a diagram showing an example of the monitoring target device (here, a milling machine). The milling machine 210 of this figure has a motor 211 and bearings 212 to 215. An accelerometer 216 and a thermocouple 217 are attached to the bearings 212 to 215, respectively. In addition, a sensor node (not shown) mounted with the above-described AI chip 10 constantly measures vibration generated during operation of the milling machine 210 to determine whether or not an abnormality occurs in the bearing.

FIG. 6 is a diagram showing a contrast example of the anomaly detection operation by new and old AI algorithms (each represented as “light” algorithm and “heavy” algorithm in the figure). On the left side of the figure, waveforms (one hour and 88 hours after) of vibration data obtained at the sensor node are depicted. On the other hand, in the upper right part of the figure, a time chart showing an abnormality detection operation by the new AI algorithm is depicted. The vertical axis indicates the distance to the hyperplane, and the horizontal axis indicates the operation time. Further, in the lower right side of the drawing, a time chart showing an abnormality detection operation by the old AI algorithm is depicted. The vertical axis indicates class output (0 = normal, 1 = abnormal), and the horizontal axis indicates operating time.

The various parameters of the new AI algorithm are as follows.
-NASA IMS data ch1 (test 2)
-Use of OCSVM-Kernel: Gauss-Gamma = 0.1
・ Data for learning: The first 200 data (33h worth of good product data is used for learning)
-Number of support vectors: 52
-Vector dimension: 99 dimensions-Feature quantity: RMS of FFT amplitude obtained every 50 Hz up to -10 kHz

Moreover, various parameters of the old AI algorithm are as follows.
-NASA IMS data-Using AlexNet (11 layer CNN)
・ Learning data Normal = 80 data (randomly selected up to 88 h)
Abnormality = 80 data (randomly selected after 88h)
・ Mini batch size: 50
・ The number of learning times: 1500

As shown in the figure, the waveforms of the vibration data obtained at the sensor node are almost identical in appearance both after one hour and after 88 hours. However, if an abnormality determination operation is performed on such vibration data by the new AI algorithm, an abnormality can be detected when the operating time of the milling machine 210 reaches 88 hours. In view of the fact that the failure actually occurred when the operating time of the milling machine 210 reached 162 hours, it can be said that the above-mentioned abnormality determination operation is extremely useful in knowing the sign of failure.

On the other hand, even by the abnormality determination operation by the old AI algorithm, an abnormality is first detected after 88 hours from the start of operation, and an abnormality is completely detected after 91 hours.

As described above, as apparent from the comparison between the two, the adoption of the new AI algorithm reduces the operation load compared to the old AI algorithm, and at the same or higher accuracy than this, the abnormality of the monitoring target device It is possible to detect Therefore, it can be said that the new AI algorithm is suitable for implementation in a sensor node requiring low power consumption and area saving.

<One class (u) -SVM using band pass filter algorithm>
When performing an anomaly detection operation for a machine health monitoring system, (a) almost all are OK data and NG data is very rare, and (b) vibration data is useful for anomaly detection of a monitored device It should be noted enough about the feature of showing information. In the following, in view of the above features, we propose an unsupervised machine learning method and a simple pre-processing method.

In view of the above characteristics, one class (u) -SVM (OCSVM) using a band pass filter (BPF) can be adopted as an AI algorithm for performing an abnormality detection operation. OCSVM is known as an unsupervised anomaly detection method. Note that (u) is introduced to clarify the upper limit value of the margin error and the lower limit value of the ratio of the support vector to all the vectors. The BPF is a means for simply obtaining frequency information without performing a high load FFT operation for the AI chip 10.

FIG. 7 is a view showing an example of the configuration of the AI chip 10 (a configuration without the post-processing unit 13). As shown in the figure, in the AI chip 10 of this configuration example, the preprocessing unit 11 includes a plurality of band pass filters 11 a connected in parallel. Further, the classifier 12 includes a support vector storage unit 12 a and a kernel operation processing unit 12 b.

A plurality of preprocessing units 11 are provided for every predetermined bandwidth Δf (Δf = 50 Hz in this figure) from the lowest frequency fmin (= DC) of input data (= raw data) to the highest frequency fmax (= 1 kHz) The band pass filter 11a is operated. These band pass filters 11a output feature values x ₀ to x _k , respectively. The preprocessing unit 11 generates a feature vector X (x ₀ , x ₁ ,..., X _k ) having the feature values x ₀ to x _k as elements, and outputs the feature vector to the classifier 12.

In the classifier 12 (OCSVM), the support vector storage unit 12 a supports the following equation (2) to construct a hyperplane (= OK / NG determination boundary) for evaluating the feature vector X. The vector X _n is stored.

Note that the above support vector Xn is a server that has an arithmetic processing capability higher than that of the AI chip 10 based on input data (= raw raw data of good product) obtained within a predetermined period after the monitoring target device is started for the first time. It is desirable to calculate.

In that case, the AI chip 10 sends input data (= good raw data) necessary for support vector calculation processing to the server before starting its normal operation (= error determination operation of input data), and the server supports We have to receive the vector X _n .

Further, the kernel operation processing unit 12b receives the input of the feature vector X and the support vector X _n , and obtains a function value f (X) using a predetermined kernel function K (see the above-mentioned equation (1)) ). Then, after the calculation, the AI chip 10 outputs the function value f (X) (= 0 to 1) as the evaluation result of the input data.

When the post-processing unit 13 not shown in the figure is included, the above function value f (X) is checked to determine whether the abnormal state should be transmitted to the server.

FIG. 8 is a diagram showing an example of the configuration of the kernel arithmetic processing unit 12b. The kernel arithmetic processing unit 12b of this configuration example includes a plurality of vector operators b10 and an adder b20.

The vector operator b10 generates a function value α _i K (X, X _i ) using the feature vector X, the support vector X _i and the coefficient α _i , respectively (where i = 0, 1,..., k).

The adder b20 generates a function value f (X) by adding together a plurality of function values α _i K (X, X _i ).

FIG. 9 is a diagram showing an exemplary configuration of the vector computing unit b10. The vector computing unit b10 of this configuration example includes a kernel computing unit b11 and a multiplier b12.

The kernel computing unit b11 receives the input of the feature vector X and the support vector X _i , and generates a function value K (X, X _i ) using a predetermined kernel function K.

There are several options for kernel function K, such as linear kernel, Gaussian kernel, or RBF kernel. For example, the linear kernel is a simple multiplication of the feature vector X and the support vector Xi, and can be implemented on the AI chip 10 with a very simple hardware configuration. On the other hand, the RBF kernel can freely define the boundary of the OK / NG determination. The following equations (3a) to (3c) are arithmetic equations of linear kernel, Gaussian kernel, and RBF kernel, respectively.

The multiplier b12 generates a function value α _i K (X, X _i ) by multiplying the function value K (X, X _i ) by the coefficient α _i .

<Sensor node>
FIG. 10 is a view showing an example of the configuration of a sensor node on which the AI chip 10 is mounted. The sensor node 1 of this configuration example functions as one component of the machine health monitoring system 300 together with the server 2, and in addition to the AI chip 10 described above, the sensor 20, the communication unit 30, An environmental power generation unit 40, a storage unit 50, and a power management unit 60 are included.

The sensor node 1 may be understood as a sensor itself attached to a monitored device (not shown) or may be understood as a gateway connected to the sensor. That is, the sensor 20 may be externally attached to the sensor node 1.

The AI chip 10 is a semiconductor device that receives power supply from the power management unit 60 and operates at the edge of the sensor 20, receives input data from the sensor 20, performs abnormality detection processing, and transmits the detection result via the communication unit 30. Report to server 2 at. Communication between the AI chip 10 and the communication unit 30 may be performed, for example, via a UART (universal asynchronous receiver / transmitter) interface. The configuration and operation of the AI chip 10 are as described above, and thus redundant description will be omitted.

The sensor 20 is a unit that receives power supply from the power management unit 60 and measures a predetermined measurement target (such as vibration or current). For example, a vibration sensor can be suitably used. In order to eliminate the high speed clock of the AI chip 10, it is desirable to use the sensor 20 as an analog output type. The most advanced vibration sensors are those of logic output type with a logic interface. The reason is that the analog output type is susceptible to noise and fatal to a high precision sensor. Therefore, when using the analog output type sensor 20, it is important to dispose the AI chip 10 in the vicinity of the sensor 20 so as not to be affected by noise.

The communication unit 30 is a module for receiving power supply from the power management unit 60 and performing wireless communication with the server 2. The AI chip 10 normally communicates with the server 2 only when error data is detected, but when sending learning data to the server 2, it is necessary to communicate with a larger capacity than normal. In view of this, it can be said that, for example, it is desirable to adopt a Wi-SUN module capable of high-speed wireless communication as the communication unit 30.

The environmental power generation unit 40 is a means (so-called energy harvester) which receives power (= vibration, light, heat, etc.) existing in the environment where the sensor node 1 is placed to generate power. When vibration is used as an energy source, a piezoelectric element such as a piezoelectric element may be used as the power generation element. In the case where sunlight or illumination light is used as an energy source, it is preferable to use a silicon-based, compound-based, or organic-based photoelectric element as the power generation element. When heat is used as an energy source, a thermoelectric element such as a Peltier element may be used as the power generation element.

In the sensor node 1, it is preferable that the measurement target of the sensor 20 and the energy source of the environmental power generation unit 40 be common. As one example, there is a case where vibration is to be measured by the sensor 20 and the above vibration is used as an energy source by the environmental power generation unit 40. In this case, when the sensor 20 tries to measure the vibration, the energy is generated by the environmental power generation unit 40 in response to the vibration, and therefore, the power to the sensor 20 can be more reliably than in the case where the energy source It becomes possible to supply.

Power storage unit 50 is a means for storing the generated power (about 100 μW) of environmental power generation unit 40, and, for example, a super capacitor (= generic name of electric double layer capacitor) can be suitably used. In particular, in order to consume a large amount of power and send a large amount of data to the server 2, a large capacity (about 1 F) supercapacitor is required as the storage unit 50.

The power management unit 60 supplies power to each unit (the AI chip 10, the sensor 20, and the communication unit 30) of the sensor node 1 using the generated power of the environmental power generation unit 40 or the stored power of the storage unit 50. It is an internal power supply circuit (for example, a DC / DC converter with a DC 3.3 V output). The environmental power generation unit 40 can not stably supply the generated power. Therefore, in order to realize stable operation of the sensor node 1, the operation of the power management unit 60 is very important. That is, in the power management unit 60, not only the storage control of the storage unit 50, but also the appropriate impedance matching control needs to be performed so that the maximum power can be obtained from the environmental power generation unit 40.

In the case of the sensor node 1 of this configuration example, since the power consumption is covered by the environmental power generation, the laying of the power supply wiring and the replacement of the battery become unnecessary. Further, since wireless communication is performed between the sensor node 1 and the server 2, signal wiring connecting the two is also unnecessary. Therefore, the sensor node 1 can be disposed at an arbitrary position.

When the server 2 receives an abnormality flag from the sensor node 1, the server 2 notifies the staff in the head office of an abnormal state. By constructing such a machine health monitoring system 300, facility maintenance can be performed by the CBM method (FIG. 2).

<Other Modifications>
In the above embodiment, the machine health monitoring system is described as an example, but the application target of the artificial intelligence algorithm (or the artificial intelligence chip mounted with the same) is not limited to this. It is possible to apply to a living body health monitoring system for managing physical condition of

Thus, various technical features disclosed in the present specification can be modified in various ways without departing from the spirit of the technical creation in addition to the above embodiment. That is, the above embodiment should be considered as illustrative in all points and not restrictive, and the technical scope of the present invention is not limited to the above embodiment, and the claims It should be understood that the scope and the meaning and meaning of the scope and all the modifications that fall within the scope are included.

The invention disclosed herein can be used, for example, in a machine health monitoring system for smart factories.

1 sensor node 10 artificial intelligence chip (AI chip)
11 pre-processing unit 11a band pass filter 12 classifier (OCSVM)
12a support vector storage unit 12b kernel operation processing unit b10 vector operation unit b11 kernel operation unit b12 multiplier b20 adder 13 post-processing unit 20 sensor 30 communication unit (Wi-SUN)
40 Environmental Power Generation Unit 50 Power Storage Unit (Super Capacitor)
60 Power Management Unit 100 Headquarters 200 Factory 210 Milling Machine 211 Motor 212-215 Bearing 216 Accelerometer 217 Thermocouple 300 Machine Health Monitoring System

Claims

Generating a feature vector by extracting feature quantities for each frequency band from input data using a plurality of band pass filters connected in parallel;
Determining a value of a kernel function using the feature vector and the support vector;
Artificial intelligence algorithm characterized by having.
The artificial intelligence algorithm according to claim 1, further comprising the step of detecting an abnormality of the input data from the value of the kernel function.
The artificial intelligence algorithm according to claim 1 or 2, wherein the input data is vibration data.
The artificial intelligence algorithm according to any one of claims 1 to 3, wherein the kernel function is a linear kernel, a Gaussian kernel, or an RBF [radial base function] kernel.
A preprocessing unit that generates a feature vector by extracting feature quantities for each frequency band from input data using a plurality of band pass filters connected in parallel;
A classifier for determining a value of a kernel function using the feature vector and the support vector;
An artificial intelligence chip characterized by having.
The artificial intelligence chip according to claim 5, further comprising a post-processing unit that performs abnormality detection of the input data from the value of the kernel function.
The artificial intelligence chip according to claim 5 or 6, wherein the classifier is an OCSVM [one class support vector machine] configured by hardware.
Sensor,
The artificial intelligence chip according to any one of claims 5 to 7, which receives the input data from the sensor.
A communication unit for performing wireless communication between the artificial intelligence chip and the server;
The sensor node characterized by having.
The sensor node according to claim 8, wherein the sensor is a vibration sensor.
Environmental power generation department,
A power storage unit for storing power generated by the environmental power generation unit;
A power management unit that supplies power to each part of a sensor node using the generated power or stored power of the storage unit;
The sensor node according to claim 8 or 9, further comprising:
11. The sensor node according to claim 10, which is attached to a monitored device.
A server that receives an abnormality flag from the sensor node;
A machine health monitoring system characterized by having:
The machine health monitoring system according to claim 11, wherein the server receives the abnormality flag from the sensor node and reports an abnormal state.