CN116681350A

CN116681350A - Intelligent factory fault detection method and system

Info

Publication number: CN116681350A
Application number: CN202310784040.9A
Authority: CN
Inventors: 刘磊
Original assignee: Nantong Beta Electric Co ltd
Current assignee: Nantong Beta Electric Co ltd
Priority date: 2023-06-28
Filing date: 2023-06-28
Publication date: 2023-09-01

Abstract

The invention provides an intelligent factory fault detection method and system, and relates to the technical field of artificial intelligence. In the invention, information comparison analysis operation is carried out between an exemplary device operation text and a device state global feature representation, and a device state global loss parameter is output; performing information-related analysis operation on the exemplary equipment monitoring video and the equipment state global feature representation, and outputting an equipment running state estimation result; analyzing an operation state evaluation loss parameter based on the device operation state estimation result and the device operation state identification; and carrying out network optimization operation on the initial operation state analysis network based on the equipment state global loss parameter and the operation state evaluation loss parameter to form a target operation state analysis network, analyzing the equipment of the to-be-processed factory through the target operation state analysis network, and outputting an equipment operation state estimation result. Based on the above, the reliability of the fault detection can be improved to some extent.

Description

Intelligent factory fault detection method and system

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a method and a system for detecting faults of an intelligent factory.

Background

The intelligent factory is a new stage of informatization development of modern factories, and information management and service are enhanced by utilizing the technology of the Internet of things and the equipment monitoring technology on the basis of the digital factory; clearly grasp the production and marketing flow, improve the controllability of the production process, reduce the manual intervention on the production line, timely and correctly collect the production line data, and reasonably schedule and schedule the production. And the novel technologies such as a green intelligent means and an intelligent system are integrated, so that a humanized factory which is efficient, energy-saving, environment-friendly and comfortable in environment is constructed. Is the result of the practical application of IBM "smart earth" concepts in manufacturing.

Artificial intelligence (Artificial Intelligence, AI for short) is a theory, method, technique and application system that simulates, extends and extends human intelligence, senses environment, obtains knowledge and uses knowledge to obtain optimal results using digital computers or digital computer controlled computations.

In the production management of the smart factory, it is necessary to perform fault detection (i.e., perform operation state analysis of the equipment) on the factory equipment in the smart factory, for example, the operation data of the factory equipment may be analyzed based on an artificial intelligence technology, however, in the prior art, there is a problem that the reliability of the fault detection is not high in the process of performing the operation data analysis.

Disclosure of Invention

Accordingly, the present invention is directed to a method and a system for intelligent factory fault detection, which can improve the reliability of fault detection to a certain extent.

In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme:

an intelligent factory fault detection method, comprising:

extracting an example device running text in a running log dimension and an example device monitoring video in a video monitoring dimension, wherein the running log dimension and the video monitoring dimension are used for carrying out network optimization operation on a first number of factory devices, the example device running text in the running log dimension and the example device monitoring video in the video monitoring dimension carry device running critical information of the first number of factory devices in different dimensions, the example device running text in the running log dimension and the example device monitoring video in the video monitoring dimension both have corresponding device running state identifiers of the factory devices, and the number of the example device running texts in the running log dimension is larger than that of the example device monitoring videos in the video monitoring dimension;

Performing information comparison analysis operation between an exemplary device operation text in the operation log dimension and device state global feature representations corresponding to the first number of types of plant devices through an initial operation state analysis network so as to output device state global loss parameters corresponding to the device state global feature representations, wherein the initial operation state analysis network carries the device state global feature representations corresponding to the first number of types of plant devices;

performing information-related analysis operation on the exemplary device monitoring video in the video monitoring dimension and the device state global feature representation corresponding to each plant device through the initial running state analysis network so as to output a device running state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension;

analyzing corresponding operation state evaluation loss parameters based on an equipment operation state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension and equipment operation state identification of the exemplary equipment monitoring video in the video monitoring dimension;

and carrying out network optimization operation on the initial operation state analysis network based on the equipment state global loss parameter and the operation state evaluation loss parameter to form a corresponding target operation state analysis network, and analyzing equipment operation data of the equipment of the to-be-processed plant through the target operation state analysis network to output a corresponding equipment operation state estimation result, wherein the equipment operation state estimation result is used for reflecting whether the equipment of the to-be-processed plant has operation faults or not, and the equipment operation data at least comprises equipment monitoring videos.

In some preferred embodiments, in the above intelligent plant fault detection method, the operation log dimension includes a second number of exemplary plant operation texts of a-th plant, the second number being smaller than the first number, and the a-th plant belongs to any plant of the first number of plant; the step of determining the global characteristic representation of the equipment state corresponding to each plant equipment in the initial running state analysis network comprises the following steps:

determining a third number of example device run texts from the second number of example device run texts, the third number being less than the second number;

a third number of device status global feature representations of the a-th plant device is determined based on the second number of example device run texts and the third number of example device run texts.

In some preferred embodiments, in the foregoing smart plant fault detection method, the step of determining a third number of device state global feature representations of the a-th plant device based on the second number of exemplary device operation texts and the third number of exemplary device operation texts includes:

Performing key information mining operation on the second number of the exemplary device operation texts through the initial operation state analysis network to output an exemplary operation feature representation corresponding to each of the second number of the exemplary device operation texts;

constructing an original first exemplary text cluster corresponding to each of the third number of exemplary device operation texts, wherein one of the original first exemplary text clusters comprises one of the third number of exemplary device operation texts;

marking the example device running text except the third number of example device running texts in the second number of example device running texts to be marked as candidate example device running texts;

performing an allocation operation on the plurality of candidate exemplary device operational texts based on feature representation deviation parameters between each of the plurality of candidate exemplary device operational text exemplary operational feature representations and the third number of exemplary device operational text exemplary operational feature representations to allocate to a third number of original first exemplary text clusters to form a third number of allocated first exemplary text clusters;

A third number of device state global feature representations is determined based on the third number of assigned first exemplary text clusters.

In some preferred embodiments, in the foregoing intelligent factory fault detection method, the step of determining a third number of device state global feature representations based on the third number of assigned first exemplary text clusters includes:

respectively carrying out mean value fusion operation on the exemplary operation characteristic representations of the operation texts of the exemplary devices included in each distributed first exemplary text cluster to form a fused first operation characteristic representation corresponding to each distributed first exemplary text cluster;

constructing a third number of original second exemplary text clusters corresponding to the first operation characteristic representations, wherein the current original second exemplary text clusters do not have any element;

performing distribution operation on the second number of the example device operation texts based on feature representation deviation parameters between the example operation feature representations of the second number of the example device operation texts and the third number of the fusion first operation feature representations respectively so as to be distributed into a third number of original second example text clusters to form a third number of distributed second example text clusters;

A third number of device state global feature representations is determined based on the third number of assigned second exemplary text clusters.

In some preferred embodiments, in the foregoing intelligent factory fault detection method, the step of determining a third number of device state global feature representations based on the third number of assigned second exemplary text clusters includes:

respectively carrying out mean value fusion operation on the exemplary operation feature representations of the operation texts of the exemplary devices included in each distributed second exemplary text cluster so as to output fusion second operation feature representations corresponding to each distributed second exemplary text cluster;

and marking a third number of the fused second operation characteristic representations to be marked as a third number of equipment state global characteristic representations.

In some preferred embodiments, in the intelligent plant fault detection method, the c-th exemplary device running text belongs to any one of the exemplary device running texts in the running log dimension, c is not greater than the number of the exemplary device running texts in the running log dimension, and one plant device corresponds to a plurality of device state global feature representations;

The step of performing, by the initial run state analysis network, an information comparison analysis operation between the example device run text in the run log dimension and the device state global feature representations corresponding to each of the first number of plant devices to output a device state global loss parameter relative to the device state global feature representations, includes:

performing key information mining operation on the c-th exemplary device operation text through the initial operation state analysis network to output an exemplary operation feature representation of the c-th exemplary device operation text;

marking the factory equipment corresponding to the c-th example equipment operation text to be marked as factory equipment to be analyzed;

marking the equipment state global feature representation with the maximum deviation parameter of the feature representation between the example operation feature representations of the c-th example equipment operation text in the plurality of equipment state global feature representations corresponding to the equipment of the plant to be analyzed so as to be marked as related equipment state global feature representations;

marking the plant equipment except the plant equipment to be analyzed in the first number of plant equipment to be marked as comparison plant equipment;

Marking the equipment state global feature representation with the minimum deviation parameter of the feature representation between the equipment state global feature representations corresponding to the comparison plant equipment and the c-th example equipment operation text in the equipment state global feature representations so as to mark the equipment state global feature representation as irrelevant;

and analyzing the equipment state global loss parameter relative to the equipment state global feature representation based on the related equipment state global feature representation and the non-related equipment state global feature representation.

In some preferred embodiments, in the foregoing intelligent plant fault detection method, the step of analyzing the device state global loss parameter relative to the device state global feature representation based on the correlated device state global feature representation and the uncorrelated device state global feature representation includes:

marking a feature representation deviation parameter between an exemplary run feature representation of the c-th exemplary device run text and the associated device state global feature representation to be a first feature representation deviation parameter;

marking a feature representation deviation parameter between the exemplary run feature representation of the c-th exemplary device run text and the non-correlated device state global feature representation to mark as a second feature representation deviation parameter;

A global loss of device state parameter relative to the global feature representation of device state is analyzed based on a difference parameter between the first feature representation deviation parameter and the second feature representation deviation parameter.

In some preferred embodiments, in the foregoing smart factory fault detection method, the b-th exemplary device monitoring video belongs to any one of the exemplary device monitoring videos in the video monitoring dimension, b is not greater than the number of the exemplary device monitoring videos in the video monitoring dimension, and a factory device corresponds to a plurality of device state global feature representations;

the step of performing information-related analysis operation on the device state global feature representation corresponding to each plant device and the exemplary device monitoring video in the video monitoring dimension through the initial running state analysis network to output a device running state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension includes:

performing key information mining operation on the b-th exemplary device monitoring video through the initial running state analysis network to output an exemplary running characteristic representation corresponding to the b-th exemplary device monitoring video;

Performing correlation analysis operation on the exemplary operation characteristic representation corresponding to the b-th exemplary equipment monitoring video and the equipment state global characteristic representation corresponding to each piece of plant equipment;

marking the equipment operation states of the plant equipment corresponding to the equipment state global feature representation with the minimum feature representation deviation parameter among the exemplary operation feature representations corresponding to the b-th exemplary equipment monitoring video determined by the correlation analysis so as to be marked as an equipment operation state estimation result of the plant equipment corresponding to the b-th exemplary equipment monitoring video;

and the step of analyzing the corresponding operation state evaluation loss parameter based on the device operation state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension and the device operation state identifier of the exemplary device monitoring video in the video monitoring dimension includes:

calculating a local running state evaluation loss parameter corresponding to the b-th exemplary equipment monitoring video based on the equipment running state estimation result of the b-th exemplary equipment monitoring video and the equipment running state identifier of the b-th exemplary equipment monitoring video;

And determining a corresponding operation state evaluation loss parameter based on the local operation state evaluation loss parameter of the b-th example equipment monitoring video.

In some preferred embodiments, in the foregoing intelligent plant fault detection method, the a-th plant belongs to any plant of the first number of plant, and one plant corresponds to a plurality of plant status global feature representations;

the step of performing network optimization operation on the initial operation state analysis network based on the equipment state global loss parameter and the operation state evaluation loss parameter to form a corresponding target operation state analysis network includes:

analyzing feature representation distinguishing data among a plurality of equipment state global feature representations corresponding to the a-th factory equipment through the initial running state analysis network; the characteristic representation distinguishing data is used for maximizing the distinction among a plurality of equipment state global characteristic representations corresponding to the a-th factory equipment;

and performing optimization adjustment operation on the network parameters of the initial operation state analysis network based on the characteristic representation distinguishing data, the equipment state global loss parameters and the operation state evaluation loss parameters to form a target operation state analysis network comprising the network parameters after optimization adjustment, wherein the network parameters comprise the equipment state global characteristic representation.

The embodiment of the invention also provides an intelligent factory fault detection system, which comprises a processor and a memory, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program so as to realize the intelligent factory fault detection method.

The intelligent factory fault detection method and system provided by the embodiment of the invention can firstly extract the operation text of the example equipment in the operation log dimension and the monitoring video of the example equipment in the video monitoring dimension; performing information comparison analysis operation between an exemplary device operation text in an operation log dimension and device state global feature representations corresponding to the first number of plant devices through an initial operation state analysis network so as to output device state global loss parameters corresponding to the device state global feature representations; performing information-related analysis operation on the exemplary equipment monitoring video in the video monitoring dimension and the equipment state global feature representation corresponding to each plant equipment through an initial running state analysis network so as to output an equipment running state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension; analyzing corresponding operation state evaluation loss parameters based on an equipment operation state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension and equipment operation state identification of the exemplary equipment monitoring video in the video monitoring dimension; and analyzing the equipment operation data of the equipment to be processed through the target operation state analysis network to output a corresponding equipment operation state estimation result. Based on the above, in the process of performing network optimization, the exemplary operation data of two dimensions, that is, the exemplary device operation text in the operation log dimension and the exemplary device monitoring video in the video monitoring dimension, are combined, so that the basis of network optimization can be richer, the mapping relationship learned by the formed target operation state analysis network is more reliable, therefore, when the operation state analysis is performed based on the target operation state analysis network, that is, when fault detection is performed, the reliability of fault detection can be improved to a certain extent, the defects in the prior art are overcome, in addition, the cost of network optimization can be reduced to a certain extent through the configuration that the number of the exemplary device operation texts in the operation log dimension is greater than the number of the exemplary device monitoring videos in the video monitoring dimension, and compared with the extraction of the operation log, the cost of video monitoring is higher.

In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

FIG. 1 is a block diagram of an intelligent factory fault detection system according to an embodiment of the present invention.

Fig. 2 is a flowchart illustrating steps involved in the intelligent factory fault detection method according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of each module included in the intelligent factory fault detection apparatus according to the embodiment of the present invention.

Description of the embodiments

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As shown in fig. 1, an embodiment of the present invention provides an intelligent factory fault detection system. Wherein the intelligent plant fault detection system may include a memory and a processor.

In detail, the memory and the processor are electrically connected directly or indirectly to realize transmission or interaction of data. For example, electrical connection may be made to each other via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that may exist in the form of software or firmware. The processor may be configured to execute the executable computer program stored in the memory, thereby implementing the intelligent factory fault detection method provided by the embodiment of the invention.

It is to be appreciated that in some embodiments, the Memory can be, but is not limited to, random access Memory (Random Access Memory, RAM), read Only Memory (ROM), programmable Read Only Memory (Programmable Read-Only Memory, PROM), erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), and the like.

It will be appreciated that in some specific embodiments, the processor may be a general purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), a System on Chip (SoC), etc.; but also Digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

It will be appreciated that in some embodiments, the intelligent plant fault detection system may be a server with data processing capabilities.

Referring to fig. 2, an embodiment of the present invention further provides a method for detecting an intelligent factory fault, which is applicable to the intelligent factory fault detection system. The method steps defined by the flow related to the intelligent factory fault detection method can be realized by the intelligent factory fault detection system.

The specific flow shown in fig. 2 will be described in detail.

Step S110, extracting an exemplary device running text in a running log dimension and an exemplary device monitoring video in a video monitoring dimension.

In the embodiment of the invention, the intelligent factory fault detection system can extract the example device running text in the running log dimension (for example, the running log text data can be extracted from the corresponding factory device to obtain the example device running text) and the example device monitoring video in the video monitoring dimension (for example, the example device monitoring video can be obtained by video monitoring through the monitoring device configured for the corresponding factory device). The operation log dimension and the video monitoring dimension are both used for performing network optimization operation on a first number of factory devices, the example device operation text in the operation log dimension and the example device monitoring video in the video monitoring dimension carry device operation critical information of the first number of factory devices in different dimensions, the example device operation text in the operation log dimension and the example device monitoring video in the video monitoring dimension both have corresponding device operation state identifiers, namely actual device operation state data, of the factory devices, and the number of the example device operation text in the operation log dimension is larger than that of the example device monitoring video in the video monitoring dimension.

And step S120, performing information comparison analysis operation between the example equipment operation text in the operation log dimension and the equipment state global feature representations corresponding to the first number of kinds of plant equipment through an initial operation state analysis network so as to output equipment state global loss parameters corresponding to the equipment state global feature representations.

In the embodiment of the invention, the intelligent factory fault detection system can perform information comparison analysis operation between the example device operation text in the operation log dimension and the device state global feature representation corresponding to each of the first number of factory devices through an initial operation state analysis network so as to output the device state global loss parameter relative to the device state global feature representation. The initial operating state analysis network carries device state global feature representations corresponding to the first number of plant devices respectively, wherein the device state global feature representations can be configured or randomly generated at the initial time, and are particularly not limited.

Step S130, performing information-related analysis operation on the device state global feature representation corresponding to each plant device and the exemplary device monitoring video in the video monitoring dimension through the initial running state analysis network, so as to output a device running state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension.

In the embodiment of the invention, the intelligent factory fault detection system can perform information-related analysis operation on the exemplary device monitoring video in the video monitoring dimension and the global feature representation of the device state corresponding to each factory device through the initial operation state analysis network so as to output the device operation state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension, namely, perform evaluation of the device operation state.

Step S140, based on the device operation state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension and the device operation state identifier of the exemplary device monitoring video in the video monitoring dimension, analyzing the corresponding operation state estimation loss parameter.

In the embodiment of the invention, the intelligent factory fault detection system can analyze the corresponding operation state evaluation loss parameters based on the device operation state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension and the device operation state identification (namely, the difference between the evaluation result and the actual data) of the exemplary device monitoring video in the video monitoring dimension.

Step S150, performing a network optimization operation on the initial operation state analysis network based on the device state global loss parameter and the operation state evaluation loss parameter to form a corresponding target operation state analysis network, and analyzing device operation data of the to-be-processed plant device through the target operation state analysis network to output a corresponding device operation state estimation result.

In the embodiment of the invention, the intelligent factory fault detection system can perform network optimization operation on the initial running state analysis network based on the equipment state global loss parameter and the running state evaluation loss parameter to form a corresponding target running state analysis network, and analyze the equipment running data of the factory equipment to be processed through the target running state analysis network to output a corresponding equipment running state estimation result. The equipment operation state estimation result is used for reflecting whether the to-be-processed plant equipment has operation faults, and the equipment operation data at least comprises equipment monitoring videos, namely, fault analysis is carried out based on the operation data of the video monitoring dimension.

Based on the above, because the exemplary running data of two dimensions, that is, the running text of the exemplary device in the running log dimension and the exemplary device monitoring video in the video monitoring dimension, are combined in the process of performing network optimization, the basis of network optimization can be richer, so that the mapping relation learned by the formed target running state analysis network is more reliable, therefore, when the running state analysis network is based on the target running state analysis network, that is, when fault detection is performed, the reliability of fault detection can be improved to a certain extent, thereby improving the defects in the prior art. And the device state global feature representations corresponding to various factory devices are optimized and adjusted through the dimension of the running log, and further the correlation analysis is carried out on the device state global feature representations corresponding to the example device monitoring video and the various factory devices respectively, so that the matching between the device state global feature representation and the example device monitoring video is realized, but the correlation analysis between the example device running text and the example device monitoring video is not realized, and the evaluation accuracy of the obtained target running state analysis network is higher.

It will be appreciated that, in some specific embodiments, the c-th example device running text belongs to any one example device running text in the running log dimension, c is not greater than the number of example device running texts in the running log dimension, a plant device corresponds to a plurality of device state global feature representations (in other embodiments, only one device state global feature representation may also be corresponding, and in addition, a specific form of a feature representation may refer to a vector), based on this, step S120 in the foregoing implementation, that is, performing, through an initial running state analysis network, an information comparison analysis operation between the example device running text in the running log dimension and the device state global feature representations corresponding to each of the first number of plant devices, so as to output a device state global loss parameter corresponding to the device state global feature representation may further include the following descriptions:

performing, by the initial run state analysis network, a key information mining operation on the c-th exemplary device run text to output an exemplary run feature representation of the c-th exemplary device run text, e.g., the c-th exemplary device run text may be subjected to a key information mining operation by a key information mining unit included in the initial run state analysis network, the key information mining unit may include one or more convolution kernels;

marking the equipment state global feature representation with the maximum value of the feature representation deviation parameter between the example operation feature representations of the c-th example equipment operation text in the plurality of equipment state global feature representations corresponding to the equipment of the plant to be analyzed so as to be marked as related equipment state global feature representations, wherein the feature representation deviation parameter can be the distance between pointing quantities;

and analyzing the equipment state global loss parameter relative to the equipment state global feature representation based on the related equipment state global feature representation and the non-related equipment state global feature representation, so that a local equipment state global loss parameter can be generated through the related equipment state global feature representation and the non-related equipment state global feature representation, the local equipment state global loss parameter can be used for reflecting the equipment state global loss parameter of the equipment state global feature representation, and the larger the local equipment state global loss parameter is, the larger the equipment state global loss parameter is, and the more inaccurate the equipment state global feature representation corresponding to the equipment of the current to-be-analyzed plant is. The local device state global loss parameter can enable the maximum characteristic representation deviation parameter (such as the characteristic representation deviation parameter between the related device state global characteristic representation and the c-th exemplary device operation text) of the type of the device to be analyzed to be smaller than the minimum distance (such as the characteristic representation deviation parameter between the non-related device state global characteristic representation and the c-th exemplary device operation text) of the type of the device to be analyzed and the type of the non-device to be analyzed, namely, the maximum distance between the device state global characteristic representation corresponding to the same device and the exemplary operation characteristic representation of the exemplary device operation text is ensured, and the minimum distance between the device state global characteristic representation corresponding to different devices and the exemplary operation characteristic representation of the exemplary device operation text is also ensured.

Wherein, it can be appreciated that, in some specific embodiments, the step of performing, by the initial running state analysis network, the key information mining operation on the c-th exemplary device running text to output the exemplary running feature representation of the c-th exemplary device running text may further include the following descriptions:

performing key information mining operation on the c-th exemplary device operation text through the initial operation state analysis network to output an initial text feature representation corresponding to the c-th exemplary device operation text, performing segmentation operation on the c-th exemplary device operation text, for example, dividing according to sentences or dividing according to semantic relativity among sentences to form a plurality of corresponding exemplary text fragments, and performing key information mining operation on each of the exemplary text fragments to output a corresponding text fragment feature representation;

performing at least one masking operation on the initial text feature representation to form at least one masked text feature representation, the masked feature representation parameters in each masking operation being not exactly identical;

For each text segment feature representation, performing a cascade of focus feature analysis operations on the text segment feature representation based on each of the masked text feature representations, respectively, to output at least one depth focus feature representation corresponding to the text segment feature representation, e.g., for text segment feature representation 1, performing a focus feature analysis operation on text segment feature representation 1 based on masked text feature representation 1 to output first focus feature representation 1, then performing a focus feature analysis operation on first focus feature representation 1 based on masked text feature representation 1 to output second focus feature representation 1, so that a final nth focus feature representation 1 may be performed sequentially, as one depth focus feature representation corresponding to text segment feature representation 1, and on the other hand, performing a focus feature analysis operation on text segment feature representation 1 based on masked text segment feature representation 2 to output first focus feature representation 2, then performing a focus feature analysis operation on first focus feature representation 2 based on masked text feature representation 2 to output second focus feature representation Jiao Tezheng, so that a final nth focus feature representation 1 may be sequentially performed as another depth focus feature representation 1;

For each text segment feature representation, performing an aggregation operation, such as a superposition or cascading combination operation, on the text segment feature representation and at least one depth focus feature representation corresponding to the text segment feature representation to form an aggregation feature representation corresponding to the text segment feature representation;

based on the corresponding aggregate feature representation of each text segment feature representation, determining an exemplary operational feature representation of the c-th exemplary device operational text, for example, each text segment feature representation may be directly subjected to an aggregate operation, such as a superposition or cascading combination operation, to form an exemplary operational feature representation of the c-th exemplary device operational text, or a focus feature analysis operation may be performed on each two text segment feature representations corresponding to the aggregate feature representation, and then, a result may be subjected to an aggregate operation to obtain an exemplary operational feature representation of the c-th exemplary device operational text.

It will be appreciated that in some specific embodiments, the step of analyzing the device state global loss parameter relative to the device state global feature representation based on the correlated device state global feature representation and the uncorrelated device state global feature representation may further include the following:

based on the difference parameter (i.e., the difference value) between the first feature representation deviation parameter and the second feature representation deviation parameter, a device state global loss parameter relative to the device state global feature representation is analyzed, e.g., the difference parameter corresponding to each exemplary device operation text may be summed and fused to obtain the device state global loss parameter.

It may be appreciated that, in some specific embodiments, the b-th exemplary device monitoring video belongs to any one of the exemplary device monitoring videos in the video monitoring dimension, b is not greater than the number of the exemplary device monitoring videos in the video monitoring dimension, and one plant device corresponds to a plurality of device state global feature representations, based on this, step S130 in the implementation content, that is, the step of performing, through the initial running state analysis network, an information-related analysis operation on the exemplary device monitoring video in the video monitoring dimension and the device state global feature representations corresponding to each plant device respectively, so as to output a device running state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension may further include the following descriptions:

Performing key information mining operation on the b-th exemplary device monitoring video through the initial running state analysis network, as described in the related manner, so as to output an exemplary running characteristic representation corresponding to the b-th exemplary device monitoring video;

performing correlation analysis operation on the exemplary operation feature representation corresponding to the b-th exemplary equipment monitoring video and the equipment state global feature representation corresponding to each piece of plant equipment, namely performing feature representation deviation parameter calculation between feature representations;

and marking the equipment operation states of the plant equipment corresponding to the equipment state global feature representation with the minimum deviation parameter between the characteristic representation deviation parameter corresponding to the b-th example equipment monitoring video and the example operation feature representation corresponding to the b-th example equipment monitoring video, so as to mark the equipment operation state estimation result of the plant equipment corresponding to the b-th example equipment monitoring video, namely determining the equipment operation states of the plant equipment corresponding to the equipment state global feature representation with the minimum difference and the most similarity.

It may be appreciated that, in some specific embodiments, step S140 in the foregoing implementation, that is, the step of analyzing the corresponding operation state evaluation loss parameter based on the device operation state estimation result of the device corresponding to the exemplary device monitoring video in the video monitoring dimension and the device operation state identifier of the exemplary device monitoring video in the video monitoring dimension, may further include the following description:

Calculating a local operation state evaluation loss parameter corresponding to the b-th exemplary equipment monitoring video based on an equipment operation state estimation result of the b-th exemplary equipment monitoring video and an equipment operation state identifier of the b-th exemplary equipment monitoring video, wherein the operation state evaluation loss parameter is used for reflecting the difference between the equipment operation state estimation result and the equipment operation state identifier, namely the difference between the evaluation result and actual data;

based on the local operation state evaluation loss parameters of the b-th exemplary equipment monitoring video, corresponding operation state evaluation loss parameters are determined, so that the local operation state evaluation loss parameters of each exemplary equipment monitoring video can be summed and fused, and the operation state evaluation loss parameters, namely the total operation state evaluation loss parameters, can be obtained.

It will be appreciated that, in some specific embodiments, the first plant belongs to any plant of the first number of plant, and a plant corresponds to a plurality of plant status global feature representations, based on which step S150 in the implementation, that is, the step of performing a network optimization operation on the initial running state analysis network to form a corresponding target running state analysis network based on the plant status global loss parameter and the running state evaluation loss parameter, may further include the following description:

Analyzing feature representation distinguishing data among a plurality of equipment state global feature representations corresponding to the a-th factory equipment through the initial running state analysis network; the feature representation distinguishing data is used for maximizing the distinction among the multiple device state global feature representations corresponding to the a-th type of factory device, so that the multiple device state global feature representations corresponding to the same type of factory device (such as the a-th type of factory device) can have larger difference, and redundant information among the multiple device state global feature representations corresponding to the same type of factory device can be reduced;

and performing an optimization adjustment operation on the network parameters of the initial operation state analysis network based on the feature representation distinguishing data, the equipment state global loss parameters and the operation state evaluation loss parameters to form a target operation state analysis network comprising the optimized and adjusted network parameters, wherein the network parameters comprise the equipment state global feature representation, i.e. the equipment state global feature representation is continuously subjected to the optimization adjustment, for example, the network parameters of the initial operation state analysis network can be subjected to the optimization adjustment operation along the directions of increasing the feature representation distinguishing data, reducing the equipment state global loss parameters and reducing the operation state evaluation loss parameters.

It will be appreciated that in some specific embodiments, the operation log dimension includes a second number of exemplary device operation texts of an a-th plant device, where the second number is smaller than the first number, and the a-th plant device belongs to any plant device of the first number, based on which the intelligent plant fault detection method may further include a step of determining a device status global feature representation corresponding to each of the plant devices in the initial operation status analysis network, where the step may further include the following description:

determining a third number of the example device operation texts from the second number of the example device operation texts, wherein the third number is smaller than the second number, the device state global feature representation can be configured according to the number of the device state global feature representations which can be determined as required, and in addition, the third number of the example device operation texts can be randomly determined from the second number of the example device operation texts, or alternatively, the third number of the example device operation texts can be decimated based on a certain rule;

and determining a third number of equipment state global feature representations of the a-th plant equipment based on the second number of the example equipment operation texts and the third number of the example equipment operation texts, such as performing operations of comparison analysis and the like.

It will be appreciated that in some specific embodiments, the step of determining the third number of device state global feature representations of the a-th plant device based on the second number of exemplary device run texts and the third number of exemplary device run texts may further include the following:

performing key information mining operation on the second number of exemplary device operation texts through the initial operation state analysis network to output an exemplary operation feature representation corresponding to each of the second number of exemplary device operation texts, for example, each of the exemplary device operation texts may be respectively subjected to key information mining operation through a key information mining unit included in the initial operation state analysis network;

constructing an original first exemplary text cluster corresponding to each of the third number of exemplary device running texts, wherein one of the original first exemplary text clusters comprises one of the third number of exemplary device running texts, namely for the exemplary device running text 1, the original first exemplary text cluster corresponding to the exemplary device running text 1 currently only comprises the exemplary device running text 1;

assigning the plurality of candidate exemplary device operational texts to a third number of original first exemplary text clusters based on feature representation deviation parameters between each of the plurality of candidate exemplary device operational text's exemplary operational feature representations and the third number of exemplary device operational text's exemplary operational feature representations, to form a third number of assigned first exemplary text clusters, e.g., the candidate exemplary device operational text may be assigned to an original first exemplary text cluster corresponding to an exemplary device operational text having the smallest feature representation deviation parameter;

and determining a third number of device state global feature representations based on the third number of assigned first exemplary text clusters, i.e., based on processing of reassigned exemplary device run text.

It will be appreciated that, in some specific embodiments, the step of determining the third number of device state global feature representations based on the third number of assigned first exemplary text clusters may further include the following description:

Respectively carrying out mean value fusion operation, namely mean value superposition on the example operation feature representations of the operation texts of the example equipment included in each distribution first example text cluster to form a fused first operation feature representation corresponding to each distribution first example text cluster;

assigning the second number of exemplary device operational texts to a third number of original second exemplary text clusters based on feature representation deviation parameters between each of the second number of exemplary device operational texts and the third number of fused first operational feature representations, to form a third number of assigned second exemplary text clusters, and similarly, assigning based on a principle that feature representation deviation parameters are minimum;

It will be appreciated that, in some specific embodiments, the step of determining the third number of device state global feature representations based on the third number of assigned second exemplary text clusters may further include the following description:

Respectively performing a mean value fusion operation, that is, a mean value superposition, on the exemplary operation feature representations of the operation texts of the exemplary devices included in each allocated second exemplary text cluster so as to output a fused second operation feature representation corresponding to each allocated second exemplary text cluster;

and marking the third number of fused second operation feature representations to be marked as the third number of equipment state global feature representations, that is, each of the fused second operation feature representations can be used as one equipment state global feature representation, and in other embodiments, further linear or nonlinear mapping operation can be performed on the fused second operation feature representations to obtain corresponding equipment state global feature representations.

Referring to fig. 3, an embodiment of the present invention further provides an intelligent factory fault detection device, which is applicable to the intelligent factory fault detection system. Wherein, wisdom mill trouble detection device includes:

the system comprises a running data extraction module, a running data extraction module and a video processing module, wherein the running data extraction module is used for extracting an example device running text in a running log dimension and an example device monitoring video in a video monitoring dimension, the running log dimension and the video monitoring dimension are used for carrying out network optimization operation on a first number of types of plant devices, the example device running text in the running log dimension and the example device monitoring video in the video monitoring dimension carry device running critical information of the first number of types of plant devices in different dimensions, the example device running text in the running log dimension and the example device monitoring video in the video monitoring dimension both have corresponding device running state identifiers of the plant devices, and the number of the example device running texts in the running log dimension is larger than that of the example device monitoring videos in the video monitoring dimension;

The information comparison analysis module is used for carrying out information comparison analysis operation between an exemplary equipment operation text in the operation log dimension and equipment state global feature representations corresponding to the first number of kinds of plant equipment through an initial operation state analysis network so as to output equipment state global loss parameters corresponding to the equipment state global feature representations, wherein the initial operation state analysis network carries the equipment state global feature representations corresponding to the first number of kinds of plant equipment;

the information correlation analysis module is used for carrying out information correlation analysis operation on the exemplary equipment monitoring video in the video monitoring dimension and the equipment state global feature representation corresponding to each factory equipment through the initial running state analysis network so as to output an equipment running state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension;

the evaluation loss determination module is used for analyzing corresponding operation state evaluation loss parameters based on an equipment operation state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension and equipment operation state identifiers of the exemplary equipment monitoring video in the video monitoring dimension;

The operation state estimation module is used for carrying out network optimization operation on the initial operation state analysis network based on the equipment state global loss parameter and the operation state evaluation loss parameter so as to form a corresponding target operation state analysis network, and analyzing equipment operation data of the to-be-processed plant equipment through the target operation state analysis network so as to output a corresponding equipment operation state estimation result, wherein the equipment operation state estimation result is used for reflecting whether the to-be-processed plant equipment has operation faults or not, and the equipment operation data at least comprises equipment monitoring videos.

In summary, the intelligent factory fault detection method and system provided by the invention can firstly extract the operation text of the example equipment in the operation log dimension and the monitoring video of the example equipment in the video monitoring dimension; performing information comparison analysis operation between an exemplary device operation text in an operation log dimension and device state global feature representations corresponding to the first number of plant devices through an initial operation state analysis network so as to output device state global loss parameters corresponding to the device state global feature representations; performing information-related analysis operation on the exemplary equipment monitoring video in the video monitoring dimension and the equipment state global feature representation corresponding to each plant equipment through an initial running state analysis network so as to output an equipment running state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension; analyzing corresponding operation state evaluation loss parameters based on an equipment operation state estimation result of equipment corresponding to the exemplary equipment monitoring video in the video monitoring dimension and equipment operation state identification of the exemplary equipment monitoring video in the video monitoring dimension; and analyzing the equipment operation data of the equipment to be processed through the target operation state analysis network to output a corresponding equipment operation state estimation result. Based on the above, in the process of performing network optimization, the exemplary operation data of two dimensions, that is, the exemplary device operation text in the operation log dimension and the exemplary device monitoring video in the video monitoring dimension, are combined, so that the basis of network optimization can be richer, the mapping relationship learned by the formed target operation state analysis network is more reliable, therefore, when the operation state analysis is performed based on the target operation state analysis network, that is, when fault detection is performed, the reliability of fault detection can be improved to a certain extent, the defects in the prior art are overcome, in addition, the cost of network optimization can be reduced to a certain extent through the configuration that the number of the exemplary device operation texts in the operation log dimension is greater than the number of the exemplary device monitoring videos in the video monitoring dimension, and compared with the extraction of the operation log, the cost of video monitoring is higher.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent factory fault detection method, comprising:

2. The intelligent plant fault detection method of claim 1, wherein a second number of exemplary plant run texts for an a-th plant is included in the run log dimension, the second number being smaller than the first number, and the a-th plant belongs to any one of the first number of plant; the step of determining the global characteristic representation of the equipment state corresponding to each plant equipment in the initial running state analysis network comprises the following steps:

3. The intelligent plant fault detection method of claim 2, wherein the step of determining a third number of device state global feature representations for the a-th plant device based on the second number of exemplary device run texts and the third number of exemplary device run texts comprises:

4. The intelligent plant fault detection method as claimed in claim 3, wherein said step of determining a third number of device state global feature representations based on said third number of assigned first exemplary text clusters comprises:

5. The intelligent plant fault detection method as claimed in claim 4, wherein the step of determining a third number of device state global feature representations based on the third number of assigned second exemplary text clusters comprises:

6. The intelligent plant fault detection method of claim 1, wherein a c-th example plant run text belongs to any one of the example plant run texts in the run log dimension, c is not greater than the number of example plant run texts in the run log dimension, and a plant device corresponds to a plurality of device state global feature representations;

7. The intelligent plant fault detection method as claimed in claim 6, wherein the step of analyzing the plant status global loss parameter relative to the plant status global feature representation based on the correlated plant status global feature representation and the uncorrelated plant status global feature representation comprises:

8. The intelligent plant fault detection method of claim 1, wherein a b-th exemplary device monitoring video belongs to any one of the exemplary device monitoring videos in the video monitoring dimension, b is not greater than the number of the exemplary device monitoring videos in the video monitoring dimension, and a plant device corresponds to a plurality of device state global feature representations;

9. The intelligent plant fault detection method as claimed in any one of claims 1 to 8, wherein a-th plant belongs to any one of the first number of plant, one plant corresponding to a plurality of plant state global feature representations;

10. An intelligent plant fault detection system, comprising a processor and a memory, the memory for storing a computer program, the processor for executing the computer program to implement the method of any of claims 1-9.