Multi-source heterogeneous information acquisition and fusion processing method for special equipment digital workshop

Abstract

The invention discloses a special equipment digital workshop multi-source heterogeneous information acquisition and fusion processing method, which comprises the following steps of S1, arranging a plurality of information acquisition devices accessed to a workshop network in a workshop so as to acquire operation state information; s2, uploading the collected information to an information preprocessing Agent of a server, and eliminating redundant information by using a wavelet denoising technology to obtain feature information; s3, uploading the feature information to an information fusion Agent of a server, identifying the feature information through a BP neural network model to obtain a feature-level information fusion result, performing similarity matching on an output result and data in a database and a knowledge base, if matching succeeds, calling an assembly state monitoring Agent of the server, and if the matching succeeds, monitoring the assembly state of the server. And human-computer interface display of a decision fusion analysis result is realized. According to the invention, the equipment and quality operation state of the general assembly workshop can be monitored and analyzed in real time, the availability of data is ensured, and timely and accurate information is provided for managers.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method for collecting and fusion processing multi-source heterogeneous information in a digital workshop of special equipment. Background Art

[0002] With the increasing application of advanced technologies such as the Internet of Things and wireless sensors in industrial enterprises, the multi-source heterogeneous information collection and fusion processing methods of special equipment digital workshops will become an important part of the informatization and intelligence of the manufacturing industry. At present, the data stored and processed in the special equipment assembly workshop is very complex, including a large amount of heterogeneous data generated by various types and uses of equipment in the production process. Manufacturing companies ensure production efficiency and product quality through real-time monitoring and data processing, but traditional methods face many limitations, such as inaccurate data collection, difficult data fusion processing, and low real-time and accuracy of monitoring. These problems hinder the intelligent optimization operation of special equipment assembly workshops. Summary of the invention

[0003] The purpose of the present invention is to provide a method for collecting and fusion processing multi-source heterogeneous information in a digital workshop of special equipment, comprising the following steps:

[0004] S1. Install several information collection external devices in the special equipment assembly production workshop and connect the information collection external devices to the workshop network;

[0005] S2. Collecting operation status information through an external information collection device;

[0006] S2, upload the collected running status information to the information preprocessing agent of the server, and use the wavelet denoising technology to eliminate redundant information to obtain feature information;

[0007] S3, the information fusion agent that uploads the feature information to the server;

[0008] The information fusion agent includes a feature-level agent and a decision-level agent;

[0009] The feature-level agent stores a BP neural network model;

[0010] The feature-level agent uses the BP neural network model to identify the feature information and obtain the feature-level information fusion result (Q, E, P); Q represents the assembly quality of the general assembly task, E represents the operation efficiency, and P represents the production progress;

[0011] Then, the decision-level agent is called to perform similarity matching between the output result (Q, E, P) and the data in the database and knowledge base. If the data matches, the process proceeds to step S4. If the data does not match, the process returns to step S2) to reacquire the operation status information.

[0012] S4. Call the assembly status monitoring Agent of the server to realize the human-machine interface display of the decision fusion analysis results.

[0013] Furthermore, the information collection external device includes a temperature and humidity sensor, a radio frequency identification device and a scanning device.

[0014] The temperature and humidity sensors, radio frequency identification devices and scanning devices are all connected through the network deployed in the workshop.

[0015] Furthermore, the characteristic information includes final assembly forming speed, final assembly forming flexibility, final assembly task quality, final assembly task time, final assembly task progress and the total number of assembly processes.

[0016] Furthermore, the information collection external device deployed in step S1 will collect the assembly task status information of different positions in the final assembly workshop in real time, which is recorded as SI, namely:

[0017]

[0018] Where: SI i Indicates the collected operation status information of the i-th type of special equipment assembly workshop;

[0019] In step S2, the information preprocessing agent preprocesses the initial information to obtain the characteristic information:

[0020]

[0021] Where: SPI i Indicates the operation status information of the i-th type special equipment assembly workshop after preprocessing;

[0022] In step S3, the feature-level fusion agent extracts features from the preprocessed information again, and the extracted feature information is:

[0023]

[0024] Where: CI i It represents the operation status information of the i-th type special equipment assembly workshop after feature extraction again.

[0025] Further, in step S2, the method of removing redundant information using wavelet denoising technology includes the following steps:

[0026] S2.1、Collect various types of operation status information SI i Normalization is performed to unify all the corresponding data information into the interval [0,1] to form the assembly status information SI′ of the relevant assembly task i ;

[0027] Normalized assembly status information SI′ i Carrying redundant information;

[0028] S2.2. Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), where n=1;

[0029] S2.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficient Wf(j,k) of the corresponding information;

[0030] j, k are the coefficients obtained by discretizing the scale factor a and the displacement factor b in the continuous wavelet transform WT(a,b);

[0031] S2.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition;

[0032] The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain the characteristic information SPI.

[0033] Further, establishing the BP neural network model includes the following steps:

[0034] S3.1. Collect N groups of operation status information SI i And the corresponding feature-level information fusion results (Q, E, P) are used as the initial sample set;

[0035] S3.2. Normalize the information of the initial sample set so that all corresponding data information is unified to the interval [0,1] to form the assembly status information SI′ of the relevant general assembly task i ;

[0036] Normalized assembly status information SI′ i Carrying redundant information;

[0037] Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), n=1, 2, ..., N;

[0038] S3.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficient Wf(j,k) of the corresponding information;

[0039] S3.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition;

[0040] The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain characteristic information SPI;

[0041] S3.5. Use characteristic information SPI to replace operation status information SI i , update the initial sample set, and obtain a training sample set with feature information SPI as input and feature-level information fusion result (Q, E, P) as output;

[0042] S3.6. Use the training sample set to train the BP neural network to obtain a trained BP neural network model.

[0043] Further, the normalization process is as follows:

[0044]

[0045] Where, SI i Indicates the collected operating status information of the i-th type of special equipment assembly workshop;

[0046] SI′ i Indicates the result after information normalization;

[0047] Indicates the maximum value of the i-th type of operating status information;

[0048] Indicates the minimum value of the i-th type of running status information.

[0049] Further, the wavelet coefficients are shown below:

[0050]

[0051] In the formula, ψ j,k (t) is the basic wavelet function;

[0052] The working threshold λ of the wavelet coefficients is as follows:

[0053]

[0054] Where: σ is the standard deviation of the redundant data following the normal distribution;

[0055] The information that does not meet the threshold conditions is eliminated using the following formula:

[0056]

[0057] Further, in step S3.6, the input sample IP and the output sample OP of the BP neural network algorithm are respectively set to:

[0058] IP=(CI1,CI2,…,CI i )=(ip1,ip2,…,ip i )

[0059] OP=(Q,E,P)=(op1,op2,op3)

[0060] IP and OP satisfy the following mapping relationship:

[0061]

[0062] Where: ip i represents the i-th type of operating status information, i=1, 2, ..., 6;

[0063] op i represents the output result of the i-th category, i=1, 2, 3;

[0064] h represents the number of hidden layer nodes in the BP neural network;

[0065] w qo Represents the output weight of the hidden layer node q to the hidden layer node o;

[0066] S q Represents the action function of the neural network error back propagation algorithm;

[0067] Represents the input weight coefficient from the input layer node p to the hidden layer node q;

[0068] ζ q Represents the threshold value of hidden layer node q.

[0069] The technical effect of the present invention is undoubted, and the beneficial effects of the present invention are as follows:

[0070] The present invention proposes a multi-source heterogeneous information collection and fusion processing method for a special equipment digital workshop, which solves the problem of poor real-time performance and low processing efficiency of multi-source heterogeneous data collection, and difficulty in supporting real-time transparent control of the assembly manufacturing process. Through the Multi-Agent technology, dynamic collaboration and intelligent interaction can be achieved, and the equipment and quality operation status of the assembly workshop can be monitored and analyzed in real time, which can ensure the availability of data and provide managers with timely and accurate information, thereby supporting the intelligent optimization operation of the special equipment assembly workshop. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It shows the performance index system of the operation status of the special equipment assembly workshop;

[0072] Figure 2 The implementation framework of the operation status information fusion of special equipment assembly workshop based on Multi-Agent is shown;

[0073] Figure 3 The method for collecting operation status information of a special equipment assembly workshop is shown;

[0074] Figure 4 The mathematical model of operation status information fusion of special equipment assembly workshop based on Multi-Agent is shown;

[0075] Figure 5 The comparison of production quality, operation efficiency and production progress results is shown. DETAILED DESCRIPTION

[0076] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0077] Embodiment 1:

[0078] A method for collecting and fusion processing multi-source heterogeneous information in a digital workshop of special equipment, comprising the following steps:

[0079] S1. Install several information collection external devices in the special equipment assembly production workshop and connect the information collection external devices to the workshop network;

[0080] S2. Collecting operation status information through an external information collection device;

[0081] S2, upload the collected running status information to the information preprocessing agent of the server, and use the wavelet denoising technology to eliminate redundant information to obtain feature information;

[0082] S3, the information fusion agent that uploads the feature information to the server;

[0083] The information fusion agent includes a feature-level agent and a decision-level agent;

[0084] The feature-level agent stores a BP neural network model;

[0085] The feature-level agent uses the BP neural network model to identify the feature information and obtain the feature-level information fusion result (Q, E, P); Q represents the assembly quality of the general assembly task, E represents the operation efficiency, and P represents the production progress;

[0086] Then, the decision-level agent is called to perform similarity matching between the output result (Q, E, P) and the data in the database and knowledge base. If the data matches, the process proceeds to step S4. If the data does not match, the process returns to step S2) to reacquire the operation status information.

[0087] S4. Call the assembly status monitoring Agent of the server to realize the human-machine interface display of the decision fusion analysis results.

[0088] Embodiment 2:

[0089] The main structure of this embodiment is the same as that of Embodiment 1. Furthermore, the information collection external device includes a temperature and humidity sensor, a radio frequency identification device and a scanning device.

[0090] The scanning device includes a two-dimensional code scanning device and a 3D scanning device.

[0091] The temperature and humidity sensors, radio frequency identification devices and scanning devices are all connected through the network deployed in the workshop.

[0092] Embodiment 3:

[0093] The main structure of this embodiment is the same as any one of Embodiments 1 to 2. Furthermore, the characteristic information includes final assembly forming speed, final assembly forming flexibility, final assembly task quality, final assembly task time, final assembly task progress and the total number of assembly processes.

[0094] Embodiment 4:

[0095] The main structure of this embodiment is the same as any one of Embodiments 1 to 3. Furthermore, the information collection external device deployed in step S1 collects assembly task status information at different locations of the final assembly workshop in real time, which is recorded as SI, namely:

[0096]

[0097] Where: SI i Indicates the collected operation status information of the i-th type of special equipment assembly workshop;

[0098] In step S2, the information preprocessing agent preprocesses the initial information to obtain the characteristic information:

[0099]

[0100] Where: SPI i Indicates the operation status information of the i-th type special equipment assembly workshop after preprocessing;

[0101] In step S3, the feature-level fusion agent extracts features from the preprocessed information again, and the extracted feature information is:

[0102]

[0103] Where: CI iIt represents the operation status information of the i-th type special equipment assembly workshop after feature extraction again.

[0104] Embodiment 5:

[0105] The main structure of this embodiment is the same as any one of Embodiments 1 to 4. Further, in step S2, the method of removing redundant information by using wavelet denoising technology includes the following steps:

[0106] S2.1、Collect various types of operation status information SI i Normalization is performed to unify all the corresponding data information into the interval [0,1] to form the assembly status information SI′ of the relevant assembly task i ;

[0107] Normalized assembly status information SI′ i Carrying redundant information;

[0108] S2.2. Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), where n=1;

[0109] S2.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficient Wf(j,k) of the corresponding information;

[0110] j, k are the coefficients obtained by discretizing the scale factor a and the displacement factor b in the continuous wavelet transform WT(a,b);

[0111] S2.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition;

[0112] The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain the characteristic information SPI.

[0113] Embodiment 6:

[0114] The main structure of this embodiment is the same as any one of Embodiments 1 to 5. Further, establishing a BP neural network model includes the following steps:

[0115] S3.1. Collect N groups of operation status information SI i And the corresponding feature-level information fusion results (Q, E, P) are used as the initial sample set;

[0116] S3.2. Normalize the information of the initial sample set so that all the corresponding data information is unified to the interval [0,1] to form the assembly status information SI′ of the relevant general assembly task i ;

[0117] Normalized assembly status information SI′ i Carrying redundant information;

[0118] Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), n=1, 2, ..., N;

[0119] S3.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficient Wf(j,k) of the corresponding information;

[0120] S3.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition;

[0121] The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain characteristic information SPI;

[0122] S3.5. Use characteristic information SPI to replace operation status information SI i , update the initial sample set, and obtain a training sample set with feature information SPI as input and feature-level information fusion result (Q, E, P) as output;

[0123] S3.6. Use the training sample set to train the BP neural network to obtain a trained BP neural network model.

[0124] Embodiment 7:

[0125] The main structure of this embodiment is the same as any one of Embodiments 5 to 6. Further, the normalization process is as follows:

[0126]

[0127] Where, SI i Indicates the collected operating status information of the i-th type of special equipment assembly workshop;

[0128] SI′ i Indicates the result after information normalization;

[0129] Indicates the maximum value of the i-th type of operating status information;

[0130] Indicates the minimum value of the i-th type of running status information.

[0131] Embodiment 8:

[0132] The main structure of this embodiment is the same as any one of Embodiments 5 to 7. Furthermore, the wavelet coefficients are as follows:

[0133]

[0134] In the formula, ψ j,k (t) is the basic wavelet function;

[0135] The working threshold λ of the wavelet coefficients is as follows:

[0136]

[0137] Where: σ is the standard deviation of the redundant data following the normal distribution;

[0138] The information that does not meet the threshold conditions is eliminated using the following formula:

[0139]

[0140] Embodiment 9:

[0141] The main structure of this embodiment is the same as any one of Embodiments 6 to 8. Further, in step S3.6, the input sample IP and the output sample OP of the BP neural network algorithm are respectively set to:

[0142] IP=(CI1,CI2,…,CI i )=(ip1,ip2,…,ip i )

[0143] OP=(Q,E,P)=(op1,op2,op3)

[0144] IP and OP satisfy the following mapping relationship:

[0145]

[0146] Where: ip i represents the i-th type of operating status information, i=1, 2, ..., 6;

[0147] op i represents the output result of the i-th category, i=1, 2, 3;

[0148] h represents the number of hidden layer nodes in the BP neural network;

[0149] w qo Represents the output weight of the hidden layer node q to the hidden layer node o;

[0150] S q Represents the action function of the neural network error back propagation algorithm;

[0151] Represents the input weight coefficient from the input layer node p to the hidden layer node q;

[0152] ζ q Represents the threshold value of hidden layer node q.

[0153] Embodiment 10:

[0154] The main structure of this embodiment is the same as any one of Embodiments 1 to 9. Furthermore, the database and knowledge base are derived from the Internet.

[0155] Embodiment 11:

[0156] The main structure of this embodiment is the same as any one of Embodiments 1 to 10. Furthermore, a method for collecting and fusing multi-source heterogeneous information in a digital workshop of special equipment includes:

[0157] S1: By calling the information collection agent, the wireless sensor network deployed in the special equipment assembly workshop is used to collect the assembly status characteristic information of the assembly task in a complex production environment.

[0158] S2: Call the information preprocessing Agent and use wavelet denoising technology to remove redundant information from the collected information.

[0159] S3: The information fusion agent system consists of a feature-level agent and a decision-level agent. The feature-level agent identifies and trains the normalized assembly state information through a three-layer BP neural network to obtain the feature-level information fusion result; then the decision-level agent matches and infers the result with the expert knowledge base and database information, and finally obtains the state analysis result of the assembly task.

[0160] S4: Call the assembly task assembly status monitoring Agent of the special equipment assembly workshop to realize the human-machine interface display of the decision fusion analysis results.

[0161] Further, in step S1, the pre-deployed sensors are connected to the network through wireless sensor network technology to collect assembly task status information at different locations in the final assembly workshop in real time, which is denoted as I. The specific process is as follows:

[0162] S01: The assembly status information collection agent collects and obtains the operation status information of the special equipment assembly workshop.

[0163]

[0164] S02: Information preprocessing Agent preprocesses the initial information to obtain feature information.

[0165]

[0166] S03: The feature-level fusion agent extracts features from the preprocessed information again to obtain extracted feature information.

[0167]

[0168] S04: Perform information fusion calculation through neural network to obtain fusion output (Q, C, P), where Q represents the assembly quality of the final assembly task, C represents the operating efficiency, and P represents the production progress.

[0169] Further, in step S2, the process of removing redundant information in the information by wavelet transform is as follows:

[0170] S01: The various operation status information is normalized through the algorithm, so that all the corresponding data information is unified into the [0,1] interval to form the assembly status information SI of the relevant general assembly task i .

[0171]

[0172] S02: Perform multi-scale, two-dimensional wavelet transform according to the following formula to obtain the required wavelet coefficients Wf(j,k).

[0173]

[0174] Where SI(t) is the normalized value of the special equipment assembly workshop operation status information obtained by the sensor at a certain moment, which consists of the original information o(t) and the noise information n(t). The noise information data follows the normal distribution N(0,σ 2 ), discretely sample SI(t), and obtain the information collected N times as SI(n), n=1,2…,N, and the number of samples is not less than 8; ψ j,k (t) represents the basic wavelet function, j, k represent the coefficients obtained by discretizing the scale factor a and the displacement factor b in the continuous wavelet transform WT(a,b) (where Usually a0=2,b0=1), then

[0175] S03: Calculate the working threshold of the wavelet coefficient according to the following threshold calculation method.

[0176]

[0177] S04: Eliminate the information that does not meet the threshold condition using the following formula.

[0178]

[0179] Further, in step S3, the parameter generation and training process of the BP neural network is as follows:

[0180] S01: Set the input sample node value to IP = (CI1, CI2, ..., CI6) = (ip1, ip2, ..., ip6), and the output sample to OP = (Q, C, P) = (op1, op2, op3). After the BP neural network model is used for calculation, IP and OP satisfy the following mapping relationship:

[0181]

[0182] S02: h represents the number of hidden layer nodes in the BP neural network, w ji represents the output weight of hidden layer node j to hidden layer node i, represents the input weight coefficient from input layer node k to hidden layer node j, ζ j represents the threshold value of hidden layer node j, S j Represents the action function of the neural network error back propagation algorithm. The Sigmoid function is introduced as the neural network error back propagation algorithm. It satisfies the following formula:

[0183] f(x)=(1+e -x ) -1

[0184] Embodiment 12:

[0185] The main structure of this embodiment is the same as any one of Embodiments 1 to 11. Furthermore, this method collects numerous and complex data information involved in the special equipment assembly workshop based on the multi-source heterogeneous information collection and fusion processing method of Multil-Agent, and performs system analysis, simulation and optimization according to corresponding standards, providing a solution for the multi-source information optimization processing and real-time monitoring of workshop status involved in the equipment task process of the special equipment assembly workshop. The technical method of the present invention is:

[0186] The present invention first proposes a method for collecting and fusing information from a special equipment assembly workshop based on a Multi-Agent method, establishes an operating status performance index system for the quality of the entire assembly process, assembly operation efficiency, and assembly production progress of the special equipment assembly workshop, and uses a BP neural network algorithm to perform fusing processing and analysis on the collected status performance index data, thereby obtaining the actual assembly production status of the special equipment assembly workshop, and verifies its feasibility through experiments. The implementation steps of the method include:

[0187] 1. Analysis of performance indicators of special equipment assembly workshop operation status

[0188] During the production task assembly process of the special equipment assembly workshop, the overall working performance and technical status of the special equipment assembly workshop gradually decrease due to the influence of factors such as equipment, logistics, personnel, production plan, technical parameters and working environment in the workshop. In order to ensure the safe and reliable operation of the assembly process in the assembly workshop and effectively grasp the real-time status of the assembly process in the assembly workshop, it is necessary to conduct a scientific and reasonable evaluation of the operation status of the assembly workshop. Since the operation status of the assembly workshop is not an intuitive and measurable data, it is necessary to comprehensively describe the assembly operation status of the assembly workshop through multiple performance indicators.

[0189] The present invention describes the completion quality of production tasks in the special equipment general assembly workshop, the efficiency of the operating status, and the completion status of the production progress, and comprehensively describes the completion status of the general assembly workshop tasks in combination with the characteristic parameters of the general assembly workshop assembly process that reflect these three types of performance indicators. The production task quality indicators are mainly reflected in the indicators of the workshop qualified product rate; the equipment production efficiency and capacity indicators are mainly reflected in the failure downtime rate; the production assembly progress indicators are mainly reflected in the number of units produced per unit time, which truly reflects the specific execution capabilities of the special equipment workshop work tasks. Its performance indicator system is shown in the attached figure Figure 1 shown.

[0190] (1) Quality

[0191] Since the quality of the final assembly workshop's operating status directly affects the quality of the products produced in the final assembly workshop, product quality is the core of reflecting the operating status of the special equipment final assembly workshop. When describing the quality situation, the product quality is mainly quantified by the qualified product rate of the final assembly products.

[0192] (2) Operational efficiency

[0193] Whether the normal operation of the workshop can be maintained under the specified operating environment is the basic performance indicator for measuring the operation status of the general assembly workshop. The equipment failure downtime rate in the general assembly workshop is the main quantitative indicator to describe the operation efficiency of the general assembly workshop.

[0194] (3) Production progress

[0195] From the perspective of the execution of production tasks, the progress of the final assembly workshop in completing the final assembly production tasks is also one of the important indicators reflecting the operation status of the final assembly workshop's assembly tasks. The ratio of the current number of completed final assembly tasks to the planned number of completed tasks is used as a quantitative indicator reflecting the production progress.

[0196] Based on the analysis of the performance indicators of the operating status of the special equipment assembly workshop, according to the actual operating characteristics of the special equipment assembly workshop, characteristic parameters reflecting the above three types of performance indicators are extracted, including assembly forming speed, assembly forming flexibility, assembly task quality, assembly task time, assembly task progress and the total number of assembly processes. The data information of each characteristic parameter is collected and acquired by the wireless sensor network inside the assembly workshop.

[0197] 2. Multi-source heterogeneous information collection and processing based on Multi-Agent

[0198] 1. Assembly workshop information collection and integration framework

[0199] The information fusion model of the running status of assembly tasks in the special equipment assembly workshop combines Multi-Agent. We usually call Agent an agent. It can be a software or a hardware with a certain degree of intelligence. The Multi-Agent system is a collection of multiple agents, which can cooperate with each other to perfectly complete the work goals. The main purpose of using the Multi-Agent system is to simplify complex systems and make it easier to communicate and manage between systems. The work of each agent is independent of each other, and they all work for their own tasks and goals, and there will be no conflict or impact between them. The system collection formed by the combination of multiple agents is mainly used to study larger and more complex problems that cannot be perfectly solved by a single agent. Based on the Multi-Agent system as the basic framework, the performance indicator data that affects the running status of the special equipment assembly workshop is fused and processed, and combined with information preprocessing, information fusion and other algorithms for design. The operation status information fusion of the special equipment assembly workshop based on Multi-Agent realizes the feature information framework that mainly affects the assembly task status in the complex production environment of the special equipment assembly workshop. It consists of a four-layer hierarchical structure of feature information collection agent, information preprocessing agent, information fusion agent and assembly workshop operation status monitoring agent. Its implementation idea is:

[0200] (1) By calling the information collection agent, the wireless sensor network deployed in the special equipment assembly workshop is used to collect the assembly status characteristic information of the assembly task in a complex production environment.

[0201] (2) Call the information preprocessing agent and use wavelet denoising technology to remove redundant information from the collected information.

[0202] (3) The information fusion agent consists of two parts: the feature-level agent and the decision-level agent. According to the information fusion requirements, the feature-level fusion agent is first called to normalize the processed assembly status information. The processed normalized status information is recognized by the neural network pattern recognizer. Appropriate neuron functions and weights are set at different nodes of the neural network to extract the characteristic values ​​of the assembly status information. The network is repeatedly trained through a three-layer BP neural network structure, and its weights are continuously adjusted until the expected performance is achieved. Then, the assembly status information to be detected is input to obtain the feature-level information fusion result output. At this time, the decision-level agent is called to match the output result with the information in the expert knowledge base and database for similarity, and reasoning and judgment are performed to finally obtain the assembly status analysis results of the special equipment assembly workshop assembly task.

[0203] (4) Call the assembly status monitoring agent of the special equipment assembly workshop to realize the human-machine interface display of the decision fusion analysis results.

[0204] The implementation framework of assembly status information fusion for assembly tasks in special equipment assembly workshop based on Multi-Agent is shown in the attached figure. Figure 2 shown.

[0205] ①Assembly status information collection agent

[0206] It is in the complex, dynamic and changeable assembly environment of the special equipment assembly workshop. It allows all agents that collect working status information through sensors to connect the pre-deployed corresponding sensors using wireless sensor network technology, and then collects information data parameters by setting different sensors, and collects assembly task status information of the assembly workshop in different orientation environments in real time, and sends the obtained redundant assembly status information to the preprocessing module for preprocessing, which is recorded as I.

[0207] ②Information preprocessing agent

[0208] It mainly uses wavelet denoising technology to realize the preprocessing of the collected special equipment assembly workshop operation status information. First, the wireless sensor network is to be collected, and the collected information is subjected to two-dimensional wavelet transform using wavelet denoising technology to obtain the wavelet coefficient value of the corresponding information, and then the working threshold of the wavelet coefficient is set. When the threshold is less than or equal to the wavelet coefficient, the original data value can be retained unchanged. When the threshold is greater than the wavelet coefficient, the original data value is set to zero. Finally, the processed information is continuously optimized by wavelet, so that redundant information can be removed well. (Wavelet transform is a new transform analysis method. It not only overcomes the corresponding shortcomings such as the window size not changing with frequency, but also inherits and develops the localization idea of ​​short-time Fourier transform. It is an ideal tool for signal time-frequency analysis and processing.)

[0209] ③Information fusion agent

[0210] In the information fusion agent, there are mainly two aspects of agents: feature-level agents and decision-level agents. According to the information fusion requirements, the feature-level fusion agent is first called to normalize the processed assembly state information, and the processed normalized state information is recognized by the neural network pattern recognizer. Appropriate neuron function functions and weights are set at different nodes of the neural network to extract the feature values ​​of the assembly state information, and the weights of the neural network are continuously adjusted, and repeated training is performed until the performance we expect is achieved. Then, the assembly state information to be detected is input to obtain the feature-level information fusion result output. At this time, the decision-level agent is called to match the output result with the information in the expert knowledge base and database for similarity, and reasoning and judgment are performed to finally obtain the operation status analysis results of the special equipment assembly workshop.

[0211] ④Assembly status monitoring agent

[0212] By calling the assembly status agent, the system can extract the processed and integrated information to the decision analysis system. The decision analysis system can display interfaces in different forms and dimensions, providing production personnel and management decision makers with intuitive display data and results of the operation status of the special equipment assembly workshop.

[0213] 2. Method for collecting information on the operation status of the general assembly workshop

[0214] The method for collecting operation status information of special equipment assembly workshop based on Multi-Agent is shown in the attached figure. Figure 3As shown in the figure. By optimizing the deployment of sensor network networking at resource points such as assembly equipment, assembly materials, and personnel in the special equipment assembly production workshop, the assembly status information collection agent can collect various information and data in the special equipment assembly production workshop, such as RFID electronic tag information scanned by workshop personnel, material information identified by handheld barcode scanners, and runtime data of equipment such as robots, so as to achieve accurate collection of information data such as assembly production lines and production unit equipment, supply chain materials, operators, and physical environment in the special equipment assembly production workshop. The data collected by the assembly status information collection agent can be uploaded to the information preprocessing agent through workshop networking, bus, etc.

[0215] In the traditional assembly mode of special equipment assembly workshops of manufacturing enterprises, the assembly process of a production and manufacturing instruction involves various resource information, such as assembly equipment, assembly materials, assembly finished products and semi-finished products, personnel, and other multi-source heterogeneous data in the workshop, which makes it difficult to achieve real-time and accurate information collection. Especially for the special equipment assembly workshop, the process planning route of its assembly tasks is complicated, the equipment is diverse, and it is difficult for workshop managers to control multiple information factors and variables, resulting in asynchronous update of workshop assembly site data, and it is difficult to accurately control the progress of the assembly task, and it is difficult to make real-time and accurate reasonable scheduling of assembly lines, materials, personnel, etc., which may delay the progress of the production instruction task of the assembly workshop. In view of this, this paper combines the previous research results and proposes a special equipment assembly workshop operation status information collection method based on Multi-Agent, which can collect multi-source information such as various production assembly resource status, production assembly progress and quality of the special equipment assembly workshop in a high-efficiency, high-reliability and high-quality manner, and it can achieve real-time, dynamic and accurate, laying a good foundation for the realization of intelligent control of the assembly workshop.

[0216] The devices required for this information collection method are optimized and configured through the workshop sensor network. For example, various information collection external devices such as handheld scanners, RFID, and assembly equipment operation status sensors are connected through the network network deployed in the workshop. At the same time, the above devices are connected to the server, gateway, hub and other devices at the edge layer of the workshop at a high speed, which can effectively realize the real-time collection of multi-source heterogeneous information in the assembly process of the general assembly workshop. The collected data is uploaded to the information preprocessing agent of the server, and the wavelet denoising technology is used to realize the preprocessing operation of the collected special equipment general assembly workshop operation status information, laying the foundation for the subsequent realization of the special equipment general assembly workshop operation status information fusion and workshop intelligent monitoring.

[0217] The idea of ​​the implementation method of the operation status information collection of the special equipment assembly workshop based on Multi-Agent is mainly as follows: the information collection requirements during the assembly task in the assembly workshop need to be arranged while meeting the process requirements of the assembly task. First, the method needs to arrange the sensor network in the workshop, such as various temperature and humidity sensor data, radio frequency identification devices, scanning devices, etc. collected during the operation of the assembly equipment in the assembly workshop, and access the workshop network. In this sensor network, all kinds of information such as equipment, materials, personnel information, etc. in each bottom workshop can be obtained by radio frequency identification or specific sensors and uploaded to the information preprocessing agent of the server. Then, the information collection device in the workshop sensor network is arranged to each resource point and information aggregation point of the assembly workshop to realize the real-time and accurate information collection of all resources in the assembly workshop, such as assembly equipment, assembly materials and workshop personnel. The JSON (JavaScript Object Notation, JS object notation) construction module in the assembly status information collection agent of the general assembly workshop is used to realize the format conversion and information description of the collected multi-source heterogeneous information and data, so that the complex multi-source heterogeneous information data in the general assembly workshop can be efficiently transmitted in the workshop network data link layer in the form of JSON format documents, and the network layer transmission of the workshop multi-source heterogeneous information data with high real-time and high efficiency can be carried out through the workshop sensor network networking that supports the edge gateway communication protocol. The information preprocessing agent module transmits the workshop environment information and the workshop assembly status information at runtime from the bottom workshop environment to the background server; it is prepared to realize the intelligent monitoring of the production planning of the assembly line, assembly equipment, assembly progress and process in the general assembly workshop.

[0218] 3. Mathematical model of assembly workshop operation status information fusion

[0219] From the special equipment assembly workshop operation status performance indicators and information fusion implementation framework, it can be seen that the key to the real-time monitoring of the operation status of the special equipment assembly workshop based on multi-agent information fusion is to solve the quantified special equipment assembly workshop operation status performance indicators by information fusion. The information preprocessing agent preprocesses the initial information to obtain feature information The feature-level fusion agent extracts features from the preprocessed information again to obtain the extracted feature information. Through the neural network, information fusion calculation is performed to obtain the fusion output (Q, E, P), where Q represents the assembly quality of the final assembly task, E represents the operating efficiency, and P represents the production progress. The output result is matched with the decision-level Agent, and finally the special equipment final assembly workshop operation status decision result S is output. The mathematical model of the special equipment final assembly workshop operation status information fusion is shown in the attached figure. Figure 4 As shown:

[0220] 4. Solution of the fusion processing model of the operation status information of the general assembly workshop

[0221] The BP neural network information fusion algorithm is an intelligent algorithm that connects multiple neurons according to a certain topological structure, so that it has good self-learning, distributed storage and data processing capabilities. It is widely used in the processing of fuzzy information problems that need to consider multiple factors at the same time. In order to solve the problem of information fusion of the operating status of the special equipment assembly workshop, the present invention improves and optimizes the three-layer BP neural network algorithm, and proposes a BP neural network algorithm suitable for the special equipment assembly workshop to solve this problem, which better realizes the information fusion and processing of many complex data. The specific steps of the proposed algorithm are as follows:

[0222] (1) Normalization of initial collected data

[0223] Since the dimensions of the data of the special equipment assembly workshop operation status information collected by the wireless sensor network of the special equipment assembly workshop are inconsistent, the algorithm normalizes the various operation status information so that the corresponding data information is unified to the [0,1] interval. The normalization process is as follows:

[0224]

[0225] Among them, SI i represents the collected operating status information of the i-th type of special equipment assembly workshop, SI′ i It represents the result after information normalization. represents the maximum value of the i-th type of assembly status information, Indicates the minimum value of the i-th type of assembly status information. The six types of performance index parameters collected are normalized to remove their dimensions.

[0226] (2) Information preprocessing

[0227] Normalize the previous information to form the assembly status information SI′ of the relevant assembly task i , and then perform multi-scale, two-dimensional wavelet transform to obtain the required wavelet coefficients Wf(j,k).

[0228]

[0229] Where SI(t) is the normalized value of the special equipment assembly workshop operation status information obtained by the sensor at a certain moment, which consists of the original information o(t) and the noise information n(t). The noise information data follows the normal distribution N(0,σ 2 ), discretely sample SI(t), and obtain the information collected N times as SI(n), n=1,2…,N, and the number of samples is not less than 8; ψ j,k (t) represents the basic wavelet function. Common basic wavelets include dbN, bior wavelet, etc. This paper adopts Daubechies (dbN) wavelet supporting discrete wavelet transform as the basic wavelet function; j, k represent the coefficients obtained by discretizing the scale factor a and the displacement factor b in the continuous wavelet transform WT (a, b) (where Usually a0=2,b0=1), then

[0230] Set the working threshold of the wavelet coefficient. When the threshold is less than or equal to the wavelet coefficient, the data value remains unchanged. When the threshold is greater than the wavelet coefficient, it is set to zero. The threshold calculation method is selected as follows:

[0231]

[0232] Through threshold processing, the information that does not meet the threshold conditions is eliminated, and the characteristic parameters of the remaining special equipment assembly workshop operating status information are obtained, thus achieving the purpose of removing redundant information.

[0233] (3) Neural network training

[0234] The characteristic parameters of the preprocessed assembly status information are input into the input nodes of the neural network, and the normalized quantitative values ​​of the assembly quality status, operating efficiency and production progress of the special equipment assembly workshop corresponding to it are used as the output of the neural network, and the neural network sample values ​​obtained before training are started. The input sample node value is set to IP = (CI1, CI2, ..., CI6) = (ip1, ip2, ..., ip6), and the output sample is OP = (Q, E, P) = (op1, op2, op3). After the BP neural network model is operated, IP and OP satisfy the following mapping relationship:

[0235]

[0236] Where h represents the number of hidden nodes in the BP neural network, w ji represents the output weight of hidden layer node j to hidden layer node i, represents the input weight coefficient from input layer node k to hidden layer node j, ζ j represents the threshold value of hidden layer node j, Sj Represents the action function of the neural network error back propagation algorithm. Introducing the Sigmoid function f(x) = (1+e -x ) -1 As the neural network error back propagation algorithm. The input node value of the BP neural network is 6, which is the same as the number of characteristic parameters, the number of hidden layer nodes is 6, and the number of neural network output nodes is 3, which correspond to the assembly quality, operation efficiency and production progress of the special equipment assembly workshop. The neural network is repeatedly trained until the expected performance is achieved.

[0237] (4) Result output

[0238] By inputting the characteristic parameters to be detected into the learned and trained neural network, the output results (Q, E, P) are obtained, which are compared with historical data to ultimately determine the operating status of the special equipment assembly workshop.

[0239] Embodiment 13:

[0240] The main structure of this embodiment is the same as any one of Embodiments 1 to 12. Figure 1 , Figure 1 It is a performance indicator system for the operating status of special equipment assembly workshops.

[0241] The three aspects of the completion quality of the production tasks of the special equipment assembly workshop, the operation status and the completion of the production progress are described, and the task completion status of the assembly workshop is comprehensively described in combination with the characteristic parameters of the assembly process of the assembly workshop that reflect these three types of performance indicators. The production task quality index is mainly reflected in the indicators of the workshop qualified product rate; the equipment production efficiency and capacity indicators are mainly reflected in the failure downtime rate; the production assembly progress indicators are mainly reflected in the number of units produced per unit time, and in the quality situation, it is mainly reflected in the workshop qualified product rate. Since the operation status of the assembly workshop directly affects the product quality, the qualified product rate has become the core indicator reflecting the product quality. When describing the quality situation, the product quality is mainly described quantitatively through the qualified product rate of the assembly products. In the operation efficiency, it is mainly reflected in the equipment failure downtime rate. Maintaining the normal workshop operation status under the specified operating environment is the key to measuring the basic performance of the assembly workshop. The equipment failure downtime rate is the main quantitative indicator to describe the operation efficiency. In the production progress, it is mainly reflected in the production quantity per unit time. The production progress of the final assembly workshop is quantified by the ratio of the current completed quantity to the planned completed quantity, which is an important indicator reflecting the operating status of the assembly task.

[0242] See also Figure 2 , Figure 2 It is a Multi-Agent based implementation framework for the operation status information fusion of special equipment assembly workshops.

[0243] First, through the assembly status information collection agent, in the complex, dynamic and changeable assembly environment of the special equipment assembly workshop, the preset sensors are connected to the network using wireless sensor network technology, and the assembly task status information in different orientation environments is collected in real time, and the redundant information is sent to the preprocessing module. Then, the information preprocessing agent uses wavelet denoising technology to perform two-dimensional wavelet transform, set the working threshold of the wavelet coefficient, optimize the processed information, and realize the preprocessing of the collected assembly status information. Then, the feature-level agent in the information fusion agent normalizes the assembly status information, and performs pattern recognition through neural networks, adjusts the weights and neuron function functions in the neural network, extracts eigenvalues, and continuously trains to improve performance. Finally, the trained neural network is used to perform feature-level information fusion on the assembly status information to be detected, and the results are output. The decision-level agent analyzes the fusion results, matches them with the information in the expert knowledge base and database, and obtains the operation status analysis results of the special equipment assembly workshop through reasoning and judgment. Finally, the system extracts the processed and integrated information into the decision analysis system through the assembly status monitoring agent, provides interface displays in different forms and dimensions, and provides production personnel and management decision makers with intuitive operation status display data and results.

[0244] See also Figure 3 , Figure 3 It is a method for collecting operating status information of special equipment assembly workshop.

[0245] Information collection during assembly tasks in the final assembly workshop requires information layout while meeting process requirements. First, it is necessary to arrange sensor networks in the workshop, such as temperature and humidity sensors, radio frequency identification devices, and scanning devices, and connect them to the workshop network. In the sensor network, the equipment, materials, and personnel information of the bottom-level workshop are uploaded to the information preprocessing agent of the server through radio frequency identification or sensors. Next, the information collection device is arranged at various resource points and information aggregation points in the workshop to realize real-time and accurate information collection of assembly equipment, assembly materials, and workshop personnel. The JSON construction module in the assembly status information collection agent is used to convert the format and describe the collected multi-source heterogeneous information, so that these complex information data can be efficiently transmitted in JSON format. Through the sensor network that supports the edge gateway communication protocol, high real-time and efficient network layer transmission is achieved. The information preprocessing agent module that transmits workshop environment information and assembly status information from the bottom layer to the background server is prepared for the intelligent monitoring of production planning such as assembly lines, assembly equipment, assembly progress, and processes in the final assembly workshop.

[0246] See also Figure 4 , Figure 4 It is a Multi-Agent based mathematical model for the fusion of operation status information of special equipment assembly workshop.

[0247] Assume that the assembly status information acquisition agent collects the operation status information of the special equipment assembly workshop, the information preprocessing agent preprocesses the initial information to obtain feature information, and the feature-level fusion agent extracts features from the preprocessed information again to obtain the extracted feature information. The information fusion calculation is performed through the neural network to obtain the fusion output Q, E, and P, where Q represents the assembly quality of the assembly task, E represents the operation efficiency, and P represents the production progress. The output result is matched with the decision-level agent, and finally the decision result S of the operation status of the special equipment assembly workshop is output.

[0248] See also Figure 5 , Figure 5 It is a comparison of production quality, operating efficiency, and production progress results.

[0249] The results obtained by collecting data and fusing information through the method proposed in this paper are consistent with the actual assembly production completion data results of the MES system of the special equipment assembly workshop and the production department in terms of quality, operating efficiency and production progress. It can be fully demonstrated that it is reasonable and feasible to monitor the operating status of the special equipment assembly workshop based on Multi-Agent information fusion.

[0250] In order to better illustrate the advantages of the technical solution of the present invention, the following is a further detailed description through specific implementation methods:

[0251] This embodiment discloses a method for multi-source heterogeneous information collection and fusion processing in a digital workshop of special equipment. In combination with the above research ideas, the application verification is carried out by taking the transmission assembly production task of the general assembly workshop of this special manufacturing enterprise as the object. The workshop of this enterprise has a total of 36 sets of box assembly units and production lines, 1 set of existing CNC network communication control system and MES system, and several industrial equipment sensors. The developed autonomous and controllable information system is used to obtain the state performance parameters collected by the sensors during the transmission assembly process of the special equipment general assembly workshop, record the production and assembly status of the special equipment general assembly workshop in five working days, group the obtained production and assembly status information, and use 10 data in each group as the input sample data for experimental verification. Using Matlab tools, after the data samples of the production and assembly status number are normalized and processed by wavelet denoising according to formulas (1), (2) and (3), the feature parameter information that removes redundancy and is available for verification can be finally obtained. After the model input data is obtained, it is input into the operation status information and fusion model of the special equipment assembly workshop based on the Multi-Agent method. At the same time, the production quality, assembly efficiency and planned progress information data of the manufacturing assembly task are normalized and denoised to form the characteristic parameters to form the training sample set of the BP neural network algorithm. After large-scale, long-term and repeated training, the expected performance indicators can be obtained. The final performance indicator parameter data is shown in Table 1.

[0252] Table 1 Performance index characteristic parameters after normalization and wavelet denoising

[0253]

[0254] The above characteristic parameter values ​​are repeatedly trained by BP neural network algorithm, and the results after repeated training are input into the special equipment assembly workshop operation status information fusion model based on Multi-Agent method. In this model, the real-time status of the assembly task of the current special equipment assembly workshop can be obtained by fusion algorithm calculation of the obtained data information by BP neural network algorithm, and the assembly quality, assembly operation efficiency and overall assembly progress in the current workshop assembly production process can be displayed with data. Then, data from different working days are collected for training and calculation, and the data samples of the production assembly state number are normalized, wavelet denoising and other links are processed to obtain the feature parameter information that removes redundancy and verifies the usable feature parameter information, which is input into the special equipment assembly workshop operation status information and fusion model based on Multi-Agent method. Through continuous verification, the characteristic parameters of the performance indicators to be fused and pre-processed are formed, and the corresponding data such as the fusion results after processing are formed. The data comparison is shown in Table 2.

[0255] Table 2 Characteristic data of performance indicators to be fused and fusion results

[0256]

[0257] By fusing the data collected using this method, the fused data is compared with the completion of the actual assembly tasks in the special equipment assembly workshop. Through the comparison of the result data, it can be seen that the data results of the quality of the whole assembly process, assembly operation efficiency and assembly production progress of the assembly workshop are basically consistent with the results obtained by collection and fusion, as shown in the attached figure. Figure 5 shown.

[0258] From the attached picture Figure 5 From the data comparison, it can be seen that the results obtained by collecting data and information fusion through the method of the present invention are consistent with the actual assembly production completion data results of the special equipment assembly workshop MES system and the production department. It can be fully demonstrated that it is reasonable and feasible to monitor the operating status of the special equipment assembly workshop based on Multi-Agent information fusion.

Claims

1. A method for collecting and fusion processing multi-source heterogeneous information in a digital workshop of special equipment, characterized in that: The following steps are involved: S1. Install several information collection external devices in the special equipment assembly production workshop and connect the information collection external devices to the workshop network; S2. Collecting operation status information through an external information collection device; S2, upload the collected running status information to the information preprocessing agent of the server, and use the wavelet denoising technology to eliminate redundant information to obtain feature information; S3, the information fusion agent that uploads the feature information to the server; The information fusion agent includes a feature-level agent and a decision-level agent; The feature-level agent stores a BP neural network model; The feature-level agent uses the BP neural network model to identify the feature information and obtain the feature-level information fusion result (Q, E, P); Q represents the assembly quality of the general assembly task, E represents the operation efficiency, and P represents the production progress; Then, the decision-level agent is called to perform similarity matching between the output result (Q, E, P) and the data in the database and knowledge base. If the data matches, the process proceeds to step S4. If the data does not match, the process returns to step S2) to reacquire the operation status information. S4. Call the assembly status monitoring Agent of the server to realize the human-machine interface display of the decision fusion analysis results.

2. According to claim 1, a method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop is characterized by: The information collection external device includes a temperature and humidity sensor, a radio frequency identification device and a scanning device; The temperature and humidity sensors, radio frequency identification devices and scanning devices are all connected through the network deployed in the workshop.

3. According to claim 1, a method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop is characterized by: The characteristic information includes final assembly forming speed, final assembly forming flexibility, final assembly task quality, final assembly task time, final assembly task progress and the total number of assembly processes.

4. The method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop according to claim 3 is characterized by: The information collection external device deployed in step S1 will collect the assembly task status information of different positions in the final assembly workshop in real time, which is recorded as SI, namely: Where: SI i Indicates the collected operation status information of the i-th type of special equipment assembly workshop; In step S2, the information preprocessing agent preprocesses the initial information to obtain the characteristic information: Where: SPI i Indicates the operation status information of the i-th type special equipment assembly workshop after preprocessing; In step S3, the feature-level fusion agent extracts features from the preprocessed information again, and the extracted feature information is: Where: CI i It represents the operation status information of the i-th type special equipment assembly workshop after feature extraction again.

5. The method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop according to claim 1 is characterized by: In step S2, the method of removing redundant information using wavelet denoising technology includes the following steps: S2.1、Collect various types of operation status information SI i Normalization is performed to unify all the corresponding data information into the interval [0,1] to form the assembly status information SI of the relevant general assembly task i ; Normalized assembly status information SI′ i Carrying redundant information; S2.

2. Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), where n=1; S2.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficient Wf(j,k) of the corresponding information; j, k are the coefficients obtained by discretizing the scale factor a and the displacement factor b in the continuous wavelet transform WT(a,b); S2.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition; The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain the characteristic information SPI.

6. The method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop according to claim 1 is characterized by: Establishing the BP neural network model includes the following steps: S3.

1. Collect N groups of operation status information SI i And the corresponding feature-level information fusion results (Q, E, P) are used as the initial sample set; S3.

2. Normalize the information of the initial sample set so that all corresponding data information is unified into the interval [0, 1] to form the assembly status information SI′ of the relevant general assembly task i ; Normalized assembly status information SI′ i Carrying redundant information; Assembly status information SI′ i Perform discrete sampling to obtain discrete operating state information SI(n), n=1, 2, ..., N; S3.3, the discrete operation state information SI(n) is subjected to two-dimensional wavelet transformation by wavelet denoising technology to obtain the wavelet coefficients Wf(j, k) of the corresponding information; S3.4, calculating the working threshold λ of the wavelet coefficients, and eliminating the information whose wavelet coefficients do not meet the threshold condition; The running status information value when the threshold is less than or equal to the wavelet coefficient is retained, and the running status information value when the threshold is greater than the wavelet coefficient is set to zero, so as to eliminate redundant information and obtain characteristic information SPI; S3.

5. Use characteristic information SPI to replace operation status information SI i , update the initial sample set to obtain a training sample set with feature information SPI as input and feature-level information fusion result (Q, E, P) as output; S3.

6. Use the training sample set to train the BP neural network to obtain a trained BP neural network model.

7. A method for collecting and fusing multi-source heterogeneous information in a digital workshop of special equipment according to claim 5 or 6, characterized in that: The normalization process is as follows: In the formula, SI i Indicates the collected operating status information of the i-th type of special equipment assembly workshop; SI′ i Indicates the result after information normalization; Indicates the maximum value of the i-th type of operating status information; Indicates the minimum value of the i-th type of running status information.

8. A method for collecting and fusing multi-source heterogeneous information in a digital workshop of special equipment according to claim 5 or 6, characterized in that: The wavelet coefficients are shown below: In the formula, ψ j , k(t) is the basic wavelet function; The working threshold λ of the wavelet coefficients is as follows: Where: σ is the standard deviation of the redundant data following the normal distribution; The information that does not meet the threshold conditions is eliminated using the following formula:

9. The method for multi-source heterogeneous information collection and fusion processing of a special equipment digital workshop according to claim 6 is characterized by: In step S3.6, the input sample IP and output sample OP of the BP neural network algorithm are respectively set to: IP=(CI1,CI2,...,CI i )=(ip1,ip2,...,ip i ) OP=(Q,E,P)=(op1,op2,op3) IP and OP satisfy the following mapping relationship: Where: ip i represents the i-th type of operating status information, i=1, 2, ..., 6; op i represents the output result of the i-th category, i=1, 2, 3; h represents the number of hidden layer nodes in the BP neural network; w qo Represents the output weight of the hidden layer node q to the hidden layer node o; S q Represents the action function of the neural network error back propagation algorithm; Represents the input weight coefficient from the input layer node p to the hidden layer node q; ζ q Represents the threshold value of hidden layer node q.

Images