Fault remote diagnosis method and system fusing equipment switching value monitoring data and multi-source service data

Abstract

The invention discloses a remote fault diagnosis method and system fusing equipment switching value monitoring data and multi-source service data, and relates to the field of electromechanical equipment fault diagnos.The method comprises the steps that firstly, equipment multi-dimensional monitoring data are obtained and preprocessed, and a digital dynamic simulation model is constructed; then, through text clustering and ensemble learning, a fuzzy expert system realizes dual-path fault diagnosis; an evidence conflict problem is solved by adopting an improved D-S evidence theory fusion result; and 3D holographic visual rescue guidance is realized in combination with a mixed reality technology. The method integrates multi-source data, improves the diagnosis accuracy, shortens the rescue time, effectively solves the pain points of inaccurate diagnosis and low rescue efficiency in the prior art, can accurately position faults, enables on-site rescue, guarantees the life and property safety of passengers, and is high in practicability.

Description

Technical Field

[0001] This invention belongs to the field of special equipment fault diagnosis and intelligent operation and maintenance technology, specifically involving a remote fault diagnosis method and system that integrates equipment switch quantity monitoring data, multimodal business data, and visual inspection data. Background Technology

[0002] Special equipment, as a crucial infrastructure in industrial production and daily life, encompasses a wide range of categories, including pressure vessels, lifting machinery, passenger ropeways, elevators, and specialized motor vehicles for use in factories and plants. Its operational safety and maintenance efficiency directly impact personnel safety, production stability, and the quality of public services. With the acceleration of industrialization and urbanization, the number of special equipment in use continues to grow, and its application scenarios are constantly expanding. Traditional fault diagnosis and maintenance technologies are no longer sufficient to meet the demands of large-scale, intelligent management, primarily exhibiting the following technical deficiencies: 1) Single data source and low diagnostic accuracy: Existing methods mostly rely on a single data source, such as only using IoT monitoring data or isolated historical fault records. They lack effective integration of multimodal business data (including inspection and testing, emergency response, maintenance records, complaint feedback, etc.), making it difficult to comprehensively depict the complex operating status and fault evolution of special equipment. This results in a high rate of misjudgment and high rate of missed diagnosis of fault causes, especially in the case of multiple concurrent faults, time-related faults, or cross-component related faults.

[0003] 2) Incomplete fault mechanism modeling: The lack of multi-dimensional and multi-scale dynamic real-time data acquisition and processing technology makes it impossible to establish accurate digital twin models of special equipment, making it difficult to simulate the fault evolution process under different working conditions, environmental interference, component aging and other scenarios. The fault location accuracy is limited, making it difficult to realize the transformation from "post-fault diagnosis" to "pre-fault prediction".

[0004] 3) Lack of visualization and intelligence in rescue guidance: Existing technologies mostly provide only text or simple diagrams. Special equipment has a complex structure and diverse categories, making it difficult for rescuers to quickly grasp the internal structural status and key fault locations of the equipment. In addition, the distribution of high-quality expert resources is uneven, making it difficult to provide accurate guidance remotely to the site. This results in slow emergency rescue response, low handling efficiency, and excessively long average fault handling time.

[0005] 4) Limitations of data fusion methods: Traditional data fusion methods such as DS evidence theory are prone to distortion of synthesis results when dealing with highly conflicting evidence. They cannot effectively integrate diagnostic conclusions from equipment physical condition monitoring data (such as switching quantities, vibration, temperature, etc.) and full life cycle business data. They also cannot take into account the complementarity of data from different dimensions, affecting the reliability of the final diagnostic results and the value of decision support. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for remote fault diagnosis that integrates equipment switch monitoring data and multi-source business data, thereby solving the aforementioned problems in the prior art.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following solution: The remote fault diagnosis method that integrates equipment switch monitoring data and multi-source service data includes the following steps: Step S1, Data Acquisition: Acquire multi-dimensional monitoring data of the target device, including basic device parameters, IoT monitoring data, and multi-modal regulatory business data; Step S2, Data Preprocessing and Database Construction: Clean, segment, and remove stop words from unstructured text data; normalize or standardize numerical data; establish a JSON-formatted equipment database using the equipment number assigned by the equipment emergency response platform as a unique identifier to achieve unified collection and management of multi-source data; Step S3, Digital Twin Model Construction: Based on the basic parameters of the equipment and IoT monitoring data, a multi-dimensional digital dynamic simulation model of the equipment is constructed through geometric modeling, physical model optimization and virtual-real fusion coupling to map the equipment's operating status in real time; Step S4, Business Data Fault Prediction: Based on multimodal regulatory business data, the process of "text vectorization - feature extraction - clustering and labeling - model training - fault prediction" is adopted. Through word embedding model, attention mechanism-Bi-LSTM model, DBSCAN clustering algorithm and Boosting ensemble learning method, a fault cause prediction model is constructed and the prediction result is output. Step S5, Diagnosis of switch quantity expert system: Establish a fuzzy expert system that includes a fault knowledge base, diagnostic rules and reasoning mechanism. Based on the timing characteristics and timing constraints of switch quantities, combined with the equipment safety operation logic, output diagnostic results based on IoT monitoring data. Step S6, Dual-Source Result Fusion Diagnosis: Using the improved DS evidence theory, the business data prediction results and switch quantity diagnosis results are fused at the decision level to resolve evidence conflict issues and output the final fault diagnosis results; Step S7, MR Visual Rescue Guidance: Based on mixed reality technology, a 3D holographic display model is developed using the Unity3D engine and MRTK. Combining artificial markers with natural feature points, the virtual and real fusion method is used to intuitively display fault information and provide rescue assistance decision-making and visual guidance.

[0008] In step S1, if the special equipment is a vertical elevator, the basic parameters of the equipment include the elevator car area, number of floors, number of stops, number of doors, traction sheave diameter, traction rope diameter, car mass, and counterweight mass, providing a basis for the construction of a digital twin model. The IoT monitoring data includes 13 types of switch quantity data from the equipment, collected in real time through contactor sensors, safety circuit sensors, door zone sensors, and other devices in the equipment control system. The sampling frequency is 100ms / time, and the data format is Boolean values ​​or integer status codes, ensuring the real-time nature and accuracy of the data. The sensors used for IoT monitoring data include contactor sensors, safety circuit sensors, and door zone sensors. The 13 types of switch quantity data specifically include data on the main contactor, safety circuit, car status, whether in the door zone, upper limit, button alarm, running contactor, running mode, running direction, door closed, lower limit, presence of people, and floor number.

[0009] Multimodal regulatory business data includes basic equipment information (brand, model, installation date, service life, equipment address, etc.), inspection data (inspection date, inspection agency, inspection results, list of non-conforming items, etc.), equipment emergency response platform data, historical fault data (fault ID, occurrence time, fault description, handling status, maintenance records, etc.), maintenance data (maintenance company, maintenance personnel, maintenance project list, maintenance result score, etc.), safety supervision system data, and rescue unit site data, covering the entire process of emergency response, inspection, complaints, and operation and maintenance, forming a complete business data chain.

[0010] Further optimization involves the following fields in step S2: basic equipment information (brand, model, installation date, etc.), equipment fault data (fault ID, occurrence time, etc.), equipment maintenance data (maintenance company, maintenance project list, etc.), and inspection data (inspection agency, list of non-conforming items, etc.).

[0011] Further optimization, in step S3, the construction of the digital twin model specifically includes: Step S3.1: Using multimodal and multiscale spatial data intelligent extraction technology, combined with laser scanning and image recognition technology, the three-dimensional spatial coordinates of the corresponding components of the equipment car, shaft, machine room and pit are obtained. A 1:1 equipment geometric model is generated through high-performance three-dimensional rendering technology to clearly restore the mechanical structure of the equipment.

[0012] Step S3.2: Adjust the physical property parameters of the geometric model according to the basic parameters of the equipment so that the mechanical properties and motion laws of the model are consistent with the actual equipment, forming a virtual mapping body.

[0013] Step S3.3: The acquired physical state data of the equipment is connected to the virtual mapping body and the cyber-physical system fusion method is used to form a multi-dimensional digital dynamic simulation model of the equipment.

[0014] Further optimization involves the following steps in step S4: Predicting business data faults. Step S4.1: Extract N historical fault records from the equipment emergency response platform database to obtain fault description text data, which will serve as the basic dataset for model training. N≥5000.

[0015] Step S4.2: The Skip-Gram model is used to process the fault description text. The window size is set to 5 and the word vector dimension is 300. Combined with the professional text training corpus corresponding to the device, the text is converted into computer-recognizable word vectors to capture the semantic information of the words.

[0016] Step S4.3: Construct a Bi-LSTM model based on an attention mechanism. The number of hidden layer units is set to 128. The attention mechanism adopts a 4-head multi-head attention, and the weights are calculated by scaling dot product attention. Bi-LSTM captures both forward and backward contextual information of the text. The attention mechanism assigns higher weights to key features, thereby improving the feature representation capability.

[0017] Step S4.4: Based on cosine similarity, the DBSCAN clustering algorithm is used to cluster fault texts and generate fault labels.

[0018] Step S4.5: Divide the dataset into training and test sets in an 8:2 ratio. Build a Boosting ensemble learning model based on XGBoost as the base model. Set the maximum tree depth to 6, the learning rate to 0.1, the number of iterations to 100, and the early stopping method to stop iterating if the performance on the validation set does not improve for 10 consecutive rounds. Use cost-sensitive learning and Focal Loss loss function to handle the sample imbalance problem.

[0019] Step S4.6: When a device malfunctions, input the device's basic information, historical fault records, and other features, and the model will output the fault cause prediction result in real time.

[0020] Further optimization involves constructing a fuzzy expert system based on the switch data from the IoT monitoring data and the fault mechanism in step S5, specifically as follows: Step S5.1, Fault Knowledge Base Construction: Organize the timing logic relationships of 13 types of switch quantities, and establish a fault judgment logic base in combination with equipment safety operation rules and technical inspection standards.

[0021] Step S5.2, Reasoning Mechanism Design: A state machine model is used to describe the timing changes of switch quantities. Switch quantity data is received in real time at 100ms / time, and logical reasoning is performed by matching the diagnostic rules in the knowledge base to quickly output the fault diagnosis results.

[0022] Further optimization involves step S6, which employs an improved DS evidence theory to fuse the business data prediction results with the switch quantity diagnosis results. Specifically, this includes: Step S6.1: Treat the business data prediction result and the switch quantity diagnosis result as two independent evidence bodies. Specifically, define the business data prediction result as evidence body 1 in the recognition framework U, with the confidence function BEL1 and the basic probability allocation. m 1. The diagnostic results of the switch quantity are evidence; 2. The reliability function BEL2 and the basic probability allocation. m 2.

[0023] Step S6.2: Calculate the degree of conflict between the two pieces of evidence using the conflict factor K. The calculation formula is as follows: , i , j =1,2,3,…, n ; in, A i This represents the focal element corresponding to the fuzzy expert system diagnostic result based on device switching quantities within the recognition framework U. B j This represents the focal element corresponding to the fault prediction result based on multi-source business data in the identification framework U; m 1( A i ) represents the basic probability allocation function of evidence body 1 based on switch quantities; it indicates the effect of the switch quantity diagnostic result on the first... i Fault type A i The degree of direct support, with a value range of [0,1]; m 2( B j ) represents the basic probability allocation function of evidence body 2 based on multi-source business data; it indicates the impact of the business data prediction result on the first... j Fault type B j The degree of direct support, with a value range of [0,1].

[0024] Step S6.3: Improve the classic DS synthesis rule using the mean K coefficient method. The probability of conflicting parts is weighted and distributed to each conclusion using the average support level of each conclusion. The calculation formula is as follows: , ; Where m(Φ)=0, and 1 / (1-K) is the normalization factor.

[0025] Further optimization is made to the MR visual rescue guidance in step S7, which specifically includes: Step S7.1: Based on the mixed reality glasses terminal, construct a 3D holographic display model of the device's operating status; Step S7.2: Place QR codes as manual identifiers at key locations on the equipment site (such as control panels and emergency exits), extract natural feature points of the equipment structure using the SIFT algorithm, and achieve MR 3D registration through ARKit / ARCore technology to ensure accurate integration of the virtual model with the real scene.

[0026] Step S7.3: The Web server sends the fault diagnosis results and digital twin model to the mixed reality glasses terminal (worn by on-site rescuers) and the expert terminal via the 4G / 5G network. The expert terminal can freeze the on-site scene and mark the fault location and operation instructions. The 2D markings are converted into 3D graphics by the terminal and projected into the device space. Rescuers can quickly locate the fault through holographic display and voice guidance and coordinate with experts in real time.

[0027] Step S7.4: The system automatically generates a rescue report, recording data such as fault type, rescue route and time taken, and iteratively optimizes the prediction model and rescue guidance algorithm based on rescue feedback.

[0028] A remote fault diagnosis system that integrates equipment switch quantity monitoring data and multi-source service data is used to realize the above-mentioned remote fault diagnosis method of integrating equipment switch quantity monitoring data and multi-source service data, including a sensor network, an edge computing node, a remote monitoring platform, an MR terminal and a mobile terminal; The sensor network is electrically connected to the device controller and is used to collect device switch quantity monitoring data, environmental data and operating status data. The switch quantity monitoring data includes 13 types of core switch quantity data, and the data is transmitted via an encryption protocol. The edge computing node is equipped with an RTOS operating system and communicates with the sensor network and remote monitoring platform through the qNCR+5G communication network to perform data cleaning, filtering, feature extraction and AI inference tasks, and supports dynamic switching between high-precision and lightweight AI models. The remote monitoring platform has a built-in JSON format database with the equipment number of the equipment emergency response platform as the unique key. It integrates a digital twin modeling module, a dual-source diagnosis module, and a result fusion module. The digital twin modeling module combines basic equipment parameters and real-time data to build a multi-dimensional dynamic simulation model. The dual-source diagnosis module includes a business data fault prediction model based on attention mechanism-Bi-LSTM and a switch quantity expert system with a dynamically updated knowledge base. The result fusion module uses an improved DS evidence theory to handle conflicts in dual-source diagnosis results. The MR terminal is wirelessly connected to the remote monitoring platform. It achieves virtual and real registration by fusing QR code manual identification with SIFT natural feature points, and provides 3D holographic fault display and remote annotation guidance from experts. The mobile terminal is used to issue post-rescue verification instructions, forming a closed loop of fault diagnosis-rescue-verification.

[0029] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned method for remote fault diagnosis of fused device switch quantity monitoring data and multi-source service data.

[0030] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly improved diagnostic accuracy: By integrating multi-source data and improving the DS evidence theory, the limitations of diagnosis from a single data source are overcome, resulting in improved fault diagnosis accuracy and reduced false alarm rate.

[0031] 2. Significantly improved rescue efficiency: Based on mixed reality-based visual guidance and remote expert collaboration, the average fault handling time is greatly shortened and the rescue waiting time is significantly reduced.

[0032] 3. High level of intelligence: The digital twin model enables panoramic perception of the equipment's operating status, and the machine learning model supports real-time prediction of fault causes, reducing reliance on human experience. Attached Figure Description

[0033] Figure 1 This is a flowchart of the remote fault diagnosis method that integrates elevator switch monitoring data and multi-source business data as described in this invention. Figure 2 This is a schematic diagram of the remote fault diagnosis method that integrates elevator switch monitoring data and multi-source business data as described in this invention. Figure 3 This is a 3D model of the elevator. Figure 4 An interface diagram for adjusting the physical property parameters of the geometric model; Figure 5 A multi-dimensional digital dynamic simulation model for elevators; Figure 6 The flowchart shows a fault prediction method based on text clustering and ensemble learning. Figure 7 This is a flowchart of a rescue technology based on MR terminal expert remote guidance. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Example 1: In this example, a vertical elevator will be used as a specific illustration, such as... Figure 1 , 2 As shown, the remote fault diagnosis method that integrates elevator switch monitoring data and multi-source business data includes the following steps: Step S1: Data Acquisition: Obtain multi-dimensional monitoring data of the target elevator, including basic elevator parameters, IoT monitoring data, and multi-modal regulatory business data.

[0036] The basic parameters of an elevator include car area, number of floors, number of stops, number of doors, traction sheave diameter, traction rope diameter, car mass, and counterweight mass, providing a foundation for the construction of a digital twin model.

[0037] The IoT monitoring data includes 13 types of switch data for elevators. These data are collected in real time through devices such as contactor sensors, safety circuit sensors, and door zone sensors in the elevator control system. The sampling frequency is 100ms / time, and the data format is Boolean value or integer status code to ensure the real-time performance and accuracy of the data.

[0038] The sensors used for IoT monitoring data include contactor sensors, safety circuit sensors, and door zone sensors. The 13 types of switch data specifically include data on the main contactor, safety circuit, car status, whether it is in the door zone, upper limit, button alarm, running contactor, running mode, running direction, door closed, lower limit, whether there are people, and floor number.

[0039] Multimodal monitoring business data includes basic elevator information (brand, model, installation date, service life, equipment address, etc.), inspection data (inspection date, inspection agency, inspection results, list of non-conforming items, etc.), elevator emergency response platform data, historical fault data (fault ID, occurrence time, fault description, handling status, maintenance records, etc.), maintenance data (maintenance company, maintenance personnel, maintenance project list, maintenance result score, etc.), safety supervision system data, and rescue unit site data, forming a complete business data chain.

[0040] Step S2, Data Preprocessing and Database Construction: Clean, segment, and remove stop words from unstructured text data; normalize or standardize numerical data; establish a JSON-formatted elevator database using the unique identifier assigned by the elevator emergency response platform to achieve unified collection and management of multi-source data.

[0041] The fields of the JSON format elevator database include basic elevator information (brand, model, installation date, etc.), elevator fault data (fault ID, occurrence time, etc.), elevator maintenance data (maintenance company, maintenance project list, etc.), and inspection data (inspection agency, list of non-conforming items, etc.).

[0042] Taking the Nanjing Elevator Emergency 96333 Platform as an example, the following is a sample of the fields in the JSON format elevator database: [{ Elevator No. 96333: "xxxx", "Elevator Basic Information": {Date of manufacture, location of equipment, etc.} "Elevator Fault Data": {Fault Time, Fault Cause, etc.} "Elevator Maintenance Data": {Maintenance Time, Maintenance Conclusions, etc.} ……… }).

[0043] Step S3: Digital Twin Model Construction: Based on the elevator's basic parameters and IoT monitoring data, a multi-dimensional digital dynamic simulation model of the elevator is constructed through geometric modeling, physical model optimization, and virtual-real fusion coupling to map the elevator's operating status in real time.

[0044] The construction of digital twin models specifically includes: Step S3.1: Employing multimodal, multi-scale spatial data intelligent extraction technology, combined with laser scanning and image recognition technology, the three-dimensional spatial coordinates of components such as the elevator car, shaft, machine room, and pit are obtained. A 1:1 elevator geometric model is then generated using high-performance 3D rendering technology. Figure 3 As shown, the mechanical structure of the elevator is clearly reproduced.

[0045] Step S3.2: Adjust the geometric model parameters according to the basic elevator parameters, such as... Figure 4 As shown, the physical property parameters of the geometric model are adjusted to make the mechanical properties and motion laws of the model consistent with the actual elevator, forming a virtual mapping body.

[0046] Step S3.3: Acquire elevator physical status data, such as floor display, running direction, door opening / closing signals, speed, acceleration, and position. Integrate 12 twin signal data points (main contactor, safety circuit, car status, door zone status, upper limit, button alarm, running contactor, running mode, running direction, door closed, lower limit, and whether anyone is present) into the elevator virtual mapping entity via an HTTP interface. A correct API call returns the following JSON data packet: {"errcode":0,"errmsg":"ok"}. An incorrect API call returns the following JSON data packet: {"errcode":1001,"errmsg":"invalid token"}. Through cyber-physical system fusion, a multi-dimensional digital dynamic simulation model of the elevator is formed, integrating virtual and physical elements. Figure 5 As shown.

[0047] Step S4, Business Data Fault Prediction: Based on multimodal regulatory business data, a fault cause prediction model is constructed using the process of "text vectorization - feature extraction - clustering and labeling - model training - fault prediction" through word embedding model, attention mechanism-Bi-LSTM model, DBSCAN clustering algorithm and Boosting ensemble learning method, and the prediction results are output.

[0048] like Figure 6 As shown, it specifically includes: Step S4.1: Extract historical fault records from the elevator emergency response platform database and obtain fault description text data as the basic dataset for model training.

[0049] In this embodiment, the experimental data came from the Nanjing 96333 platform, totaling 50,000 fault records.

[0050] Step S4.2: The Skip-Gram model is used to process the fault description text. The window size is set to 5 and the word vector dimension is 300. Combined with professional text training corpus in the elevator field, the text is converted into word vectors that can be recognized by computers to capture the semantic information of words.

[0051] Step S4.3: To address the issue that Long Short-Term Memory (LSTM) networks cannot utilize the contextual information of elevator fault text, this invention connects two LSTMs with opposite time sequences to the same output, forming a Bi-LSTM model. At time t, the output H(t) contains both forward and backward information. By obtaining context information of the input fault text sequence, the accuracy of the model was improved.

[0052] To allocate computational resources rationally and reduce the impact of non-critical factors, an attention mechanism is introduced into Bi-LSTM. The output of the Bi-LSTM layer... H (t (Through hidden layer) B The hidden layer state value is u t , A =[ a 1, a 2, …, a n [ ] represents the attention weights of the hidden states from the past inputs to the current input, and the final feature output. v The sum of hidden states with different attention weights is calculated as follows: u t =tanh(e multi H(t)+b); a t =softmax (u t T u w ); v =∑ t a t H(t); Where emulti is the intermediate representation of the attention score, b is the bias parameter, and u w For the context vector, a t For attention weights.

[0053] After obtaining word features using an improved Bi-LSTM, multi-level comparisons are performed at three levels: word vectors with word vectors, sentence vectors with sentence vectors, and word vectors with sentence vectors. The final similarity is then calculated using weighted averages.

[0054] Step S4.4: Cluster all elevator fault information according to similarity metric. The same fault description is clustered into one class. The cluster is used as the elevator fault label to label the elevator historical fault dataset.

[0055] In this embodiment, historical fault datasets from January 2014 to December 2024 in Nanjing City were extracted from the elevator database of the 96333 emergency platform. Elevator fault text clustering algorithms were used to calculate 12 fault categories: Fault 1: Power outage; Fault 2: Door jammed by trash; Fault 3: Automatic recovery after fault; Fault 4: Door lock malfunction; Fault 5: Overload; Fault 6: Trip; Fault 7: Safety circuit disconnection; Fault 8: Door guide system malfunction; Fault 9: Control system crash; Fault 10: Door deformation due to rough handling; Fault 11: Machine room overheating; Fault 12: No cause found by grid rescue. Statistical analysis revealed that most fault numbers exhibit a certain periodicity, with a sharp decrease in February each year, followed by a gradual increase each month. July, August, November, and December see a higher number of faults, possibly due to hot and humid weather or cold weather. However, sometimes the periodicity is not significant or data is severely missing, such as in fault categories 2 and 10, which may be due to subjective factors in the fault causes recorded by different personnel.

[0056] Step S4.5: Using the labeled historical elevator malfunction dataset, the basic information of the malfunctioning elevator data is converted into one-hot vectors or normalized. Historical malfunction information is then added to the features through normalization or a log function. The obtained features and annotation information are used to divide the training and testing sets in an 8:2 ratio. A Boosting ensemble learning model is constructed based on XGBoost, with a maximum tree depth of 6, a learning rate of 0.1, and an early stopping mechanism where iteration stops if the validation set performance shows no improvement for 10 consecutive rounds. Cost-sensitive learning and Focal Loss are used to address the sample imbalance problem.

[0057] Step S4.6: When the elevator malfunctions, input the elevator's basic information, historical fault records, and other features, and the model will output the fault cause prediction result in real time.

[0058] In this embodiment, elevator fault data training and testing sets are constructed using the obtained features and annotation information. For each fault data point, basic elevator information and historical fault information prior to the current fault are used as input features, and 12 fault categories are used as labels, resulting in a total of 44,350 data points. 80% (35,480) of the data are used as the training set, and 20% (8,870) are used as the testing set. Training is performed using Boosting ensemble learning, and early stopping is used during iteration to prevent overfitting. The results show that the optimal results are achieved after 70 iterations, with the model achieving a top-1 accuracy of 66.2% and a top-3 accuracy (predicting the three most likely fault causes) of 86.1% on the testing set. Simultaneously, comparisons are made with C4.5 decision trees, Naïve Bayes classifiers, SVM classifiers, and GBDT algorithms using 10-fold cross-validation. The experimental results are shown in Table 1. Table 1 shows that the elevator fault cause prediction algorithm described in this invention outperforms the selected comparison algorithms in all metrics.

[0059] A model for predicting elevator malfunctions was constructed based on the Boosting ensemble learning algorithm. This model predicts the cause of new elevator malfunctions based on historical malfunction records. Experimental comparisons demonstrate that using the Boosting ensemble learning algorithm to predict elevator malfunctions is feasible, especially when recommending the top-k predictions, the coverage of correct results meets practical requirements.

[0060] Table 1 Comparison of Experimental Results for Predicting Elevator Failure Causes

[0061] Step S5, Expert System Diagnosis for Switching Quantities: Establish a fuzzy expert system including a fault knowledge base, diagnostic rules, and reasoning mechanisms. Based on the timing characteristics and constraints of switching quantities, and combined with the elevator safety operation logic, output diagnostic results based on IoT monitoring data. The fuzzy expert system is constructed based on IoT switching quantity monitoring data and fault mechanisms, specifically as follows: Step S5.1, Fault Knowledge Base Construction: The timing logic relationships of 13 types of switch quantities are analyzed. Combined with elevator safety operation rules and technical inspection standards, a fault judgment logic base is established, covering judgment rules for 18 types of faults, including leveling entrapment, overshooting, and door opening timeout. The judgment rules are as follows: If the "safety circuit" is disconnected and the "car status" is not in the door zone, the diagnosis is "non-leveling entrapment". If the "upper limit" is triggered and the "running direction" is upward, it is diagnosed as "overshooting failure"; If the "door closed" signal is false and lasts for more than 10 seconds, the "door open timeout" alarm will be triggered.

[0062] Step S5.2, Reasoning Mechanism Design: A state machine model is used to describe the timing changes of switch quantities. Switch quantity data is received in real time at 100ms / time, and logical reasoning is performed by matching the diagnostic rules in the knowledge base to quickly output the fault diagnosis results.

[0063] Step S6, Dual-Source Result Fusion Diagnosis: Using a modified DS evidence theory based on the mean K coefficient method, decision-level fusion is performed on the business data prediction results and switch quantity diagnostic results to resolve evidence conflicts and output the final fault diagnosis result. The improved DS evidence theory fusion process specifically includes: Step S6.1: Treat the business data prediction result and the switch quantity diagnosis result as two independent evidence bodies. Specifically, define the business data prediction result as evidence body 1 in the recognition framework U, with the confidence function BEL1 and the basic probability allocation. m 1. The diagnostic results of the switch quantity are evidence; 2. The reliability function BEL2 and the basic probability allocation. m 2; Step S6.2: Calculate the degree of conflict between the two pieces of evidence using the conflict factor K. The calculation formula is as follows: , i , j =1,2,3,…, n ; in, A i This represents the focal element corresponding to the diagnostic result of the fuzzy expert system based on elevator switching quantities in the recognition framework U. B j This represents the focal element corresponding to the fault prediction result based on multi-source business data in the identification framework U; m 1( A i ) represents the basic probability allocation function of evidence body 1 based on switch quantities; it indicates the effect of the switch quantity diagnostic result on the first... i Fault type A i The degree of direct support, with values ​​ranging from [0,1]; for example m 1 (“overshoot fault”) = 0.8, which means that the direct confidence level of the switch data for “overshoot fault” is 80%.

[0064] m 2( B j ) represents the basic probability allocation function of evidence body 2 based on multi-source business data; it indicates the impact of the business data prediction result on the first... j Fault type B j The degree of direct support, for example m2 ("Door lock failure") = 0.7, which means that the business data has a direct trust level of 70% for "door lock failure".

[0065] Step S6.3: Improve the classic DS synthesis rule using the mean K coefficient method. The probability of conflicting parts is weighted and distributed to each conclusion using the average support level of each conclusion. The calculation formula is as follows: , ; Where 1 / (1-K) is the normalization factor.

[0066] Step S7, MR Visual Rescue Guidance: Based on mixed reality technology, a 3D holographic display model is developed using the Unity3D engine and MRTK. This model combines artificial markers with natural feature points to create a virtual-real fusion method, intuitively displaying fault information and providing rescue decision support and visual guidance. For example... Figure 7 As shown, it specifically includes: Step S7.1: Construct a 3D holographic display model of the elevator's operating status based on the mixed reality glasses terminal.

[0067] Step S7.2: Place QR codes as manual identifiers at key locations on the elevator site (such as control panels and emergency exits), extract natural feature points of the elevator structure using the SIFT algorithm, and achieve MR 3D registration using ARKit / ARCore technology to ensure accurate integration of the virtual model with the real scene.

[0068] Step S7.3: The Web server sends the fault diagnosis results and digital twin model to the mixed reality glasses terminal (worn by on-site rescue personnel) and the expert terminal (PC or Android mobile terminal) via 4G / 5G network.

[0069] Among them, the expert terminal can freeze the scene and mark the fault location and operation instructions. The 2D markings are converted into 3D graphics and projected onto the elevator space. Rescue personnel can quickly locate the fault through holographic display and voice guidance and coordinate with experts in real time to handle the situation. Step S7.4: The system automatically generates a rescue report, recording data such as fault type, rescue route, and time consumption, and iteratively optimizes the prediction model and rescue guidance algorithm based on rescue feedback.

[0070] Example 2: A remote fault diagnosis system that integrates elevator switch monitoring data and multi-source business data, including a sensor network, edge computing nodes, a remote monitoring platform, an MR terminal, and a mobile terminal.

[0071] The sensor network is electrically connected to the elevator controller and is used to collect elevator switch quantity monitoring data, environmental data and operating status data. The switch quantity monitoring data includes 13 types of core switch quantity data, and the data is transmitted via an encryption protocol. The edge computing node is equipped with an RTOS operating system and communicates with the sensor network and remote monitoring platform through the qNCR+5G communication network to perform data cleaning, filtering, feature extraction and AI inference tasks, and supports dynamic switching between high-precision and lightweight AI models.

[0072] The remote monitoring platform has a built-in JSON format database with the elevator number of the elevator emergency response platform as the unique key. It integrates a digital twin modeling module, a dual-source diagnosis module, and a result fusion module. The digital twin modeling module combines basic elevator parameters with real-time data to construct a multi-dimensional dynamic simulation model. The dual-source diagnosis module includes a business data fault prediction model based on attention mechanism-Bi-LSTM and a switch quantity expert system with a dynamically updated knowledge base. The result fusion module uses an improved DS evidence theory to handle conflicts in dual-source diagnosis results.

[0073] The MR terminal is wirelessly connected to the remote monitoring platform. It achieves virtual and real registration by fusing QR code manual identification with SIFT natural feature points, and provides 3D holographic fault display and remote annotation guidance from experts. The mobile terminal is used to issue post-rescue verification instructions, forming a closed loop of fault diagnosis-rescue-verification.

[0074] Example 3: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned method for remote fault diagnosis that integrates elevator switch monitoring data and multi-source business data.

[0075] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A fault remote diagnosis method of fusing switch quantity monitoring data and multi-source service data, characterized in that, The method comprises the following steps: Step S1, data collection: obtaining multi-dimensional monitoring data of the target device, the multi-dimensional monitoring data comprising device basic parameters, Internet of Things monitoring data and multi-modal supervision business data; Step S2, data preprocessing and database construction: cleaning, word segmentation and stop word removal processing are performed on unstructured text data, and numerical value type data is normalized or standardized; a device number allocated by an emergency disposal platform is taken as a unique identifier, a JSON format device database is established, and unified collection and management of multi-source data are realized; Step S3, digital twin model construction: according to the device basic parameters and the Internet of Things monitoring data, a device multi-dimensional digital dynamic simulation model is constructed through geometric modeling, physical model optimization and virtual-real fusion coupling, and the device running state is mapped in real time; Step S4, business data fault prediction: based on the multi-modal supervision business data, a "text vectorization-feature extraction-clustering tagging-model training-fault prediction" process is adopted, a fault cause prediction model is constructed through a word embedding model, an attention mechanism-Bi-LSTM model, a DBSCAN clustering algorithm and a Boosting integrated learning method, and a prediction result is output; Step S5, on-off quantity expert system diagnosis: a fuzzy expert system comprising a fault knowledge base, a diagnosis rule and a reasoning mechanism is established, a diagnosis result based on the Internet of Things monitoring data is output based on the on-off quantity time sequence characteristics and time sequence constraints and in combination with the device safe operation logic; Step S6, dual-source result fusion diagnosis: an improved D-S evidence theory is adopted to perform decision-level fusion on the business data prediction result and the on-off quantity diagnosis result, an evidence conflict problem is solved, and a final fault diagnosis result is output; Step S7, MR visual rescue guidance: based on a mixed reality technology, a 3D holographic display model is developed by using a Unity3D engine and an MRTK, a virtual-real fusion method combining an artificial mark and a natural feature point is adopted, and fault information is intuitively displayed and rescue auxiliary decision and visual guidance are provided.

2. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 1, characterized in that, In the step S2, the fields of the JSON format device database comprise device basic information, device fault data, device maintenance data and inspection data.

3. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 2, characterized in that, In the step S3, the digital twin model construction specifically comprises: Step S3.1, a multi-modal multi-scale spatial data intelligent extraction technology is adopted, laser scanning and image recognition technology are combined, three-dimensional spatial coordinates of device corresponding components are obtained, and a 1:1 device geometric model is generated by using a high-performance three-dimensional rendering technology; Step S3.2, the physical property parameters of the geometric model are adjusted according to the device basic parameters, the mechanical characteristics and motion law of the model are consistent with the actual device, and a virtual mapping body is formed; Step S3.3, the obtained device physical state data is connected to the virtual mapping body and an information-physical system fusion method is adopted, and a device multi-dimensional digital dynamic simulation model is formed.

4. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 3, characterized in that, In the step S4, the business data fault prediction specifically comprises: Step S4.1, N pieces of historical fault records are extracted from the device emergency disposal platform database, fault description text data is obtained, and the data is taken as a basic data set for model training, N≥5000; Step S4.2, the fault description text is processed by using a Skip-Gram model, professional text training corpus corresponding to the device is combined, the text is converted into a computer recognizable word vector, and semantic information of the words is captured; Step S4.3, a Bi-LSTM model based on an attention mechanism is constructed, the number of hidden layer units is set to 128, the attention mechanism adopts 4 multi-head attention, and the weight is calculated by scaling dot product attention; the Bi-LSTM model simultaneously captures context information in a forward direction and a backward direction of the text, the attention mechanism gives a higher weight to key features, and the feature expression capability is improved; Step S4.4, based on cosine similarity, a DBSCAN clustering algorithm is used to cluster the fault text and generate a fault label; Step S4.5, a data set is divided into a training set and a test set according to a proportion of 8:2, an XGBoost is used as a basic model to construct a Boosting integrated learning model, the maximum depth of a tree is set to 6, a learning rate is set to 0.1, the number of iterations is set to 100, an early stopping method is set to stop iteration when the performance of a verification set does not improve for 10 rounds in succession, and a cost-sensitive learning and a Focal Loss loss function are used to process a sample imbalance problem; Step S4.6, when a device fails, device basic information and historical fault record features are input, and the model outputs a fault cause prediction result in real time.

5. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 4, characterized in that, In the step S5, a fuzzy expert system is constructed based on on-off data in Internet of Things monitoring data and in combination with a fault mechanism, and specifically, Step S5.1, a fault knowledge base is constructed: time sequence logical relationships of 13 types of on-off data are combed, a fault judgment logic base is established in combination with a device safe operation rule and a technical inspection standard; Step S5.2, a reasoning mechanism is designed: a state machine model is used to describe time sequence changes of the on-off data, on-off data is received in real time, a diagnosis rule in the knowledge base is matched to perform logical reasoning, and a fault diagnosis result is quickly output.

6. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 5, characterized in that, In the step S6, an improved D-S evidence theory is used to fuse a business data prediction result and an on-off diagnosis result, and specifically, Step S6.1, the service data pre-judgment result and the switching value diagnosis result are respectively taken as two independent evidence bodies, specifically: in the definition recognition framework U, the service data pre-judgment result is evidence body 1, the belief function BEL1, the basic probability distribution m 1, and the switching value diagnosis result is evidence body 2, the belief function BEL2, the basic probability distribution m 2. Step S6.2, a conflict factor K is used to calculate a conflict degree of two evidence bodies, and a calculation formula is as follows: ， i , j =1,2,3,…, n ； Wherein, A i represents the focal element corresponding to the diagnosis result of the fuzzy expert system based on the device switch value in the identification framework U, B j represents the focal element corresponding to the fault prediction result based on multi-source business data in the identification framework U. m 1( A i ) is the basic probability assignment function of the evidence body 1 based on the switch value, which represents the direct support degree of the switch value diagnosis result to the first i A i fault type, with a value range of [0, 1]; m 2( B j ) is the basic probability assignment function of the evidence body 2 based on multi-source business data, which represents the direct support degree of the business data prediction result to the first j B j fault type, with a value range of [0, 1];​​ Step S6.3, a mean K coefficient method is used to improve a classical D-S synthesis rule, an average support degree of each conclusion is used as a weight, a probability of a conflict part is weightedly distributed to each conclusion, and a calculation formula is as follows: ， ； Wherein, m (Φ)=0, 1 / (1-K) is a normalization factor.

7. The fusion device switch quantity monitoring data and multi-source service data fault remote diagnosis method according to claim 6, characterized in that, In the step S7, the MR visual rescue guidance specifically includes: Step S7.1, a device operation state 3D holographic display model is constructed based on a mixed reality glasses terminal; Step S7.2, QR codes are arranged as artificial markers at key parts of a device site, natural feature points of a device structure are extracted by using a SIFT algorithm, MR three-dimensional registration is realized by using ARKit / ARCore technology, and accurate fusion of a virtual model and a real scene is ensured; Step S7.3, a Web server end sends a fault diagnosis result and a digital twin model to the mixed reality glasses terminal and an expert end through a 4G / 5G network; Among them, the mixed reality glasses are worn by on-site rescue personnel, and the expert end can freeze the on-site picture and mark the fault part and operation guide. The 2D mark is converted into a 3D graphic by the terminal and projected to the device space. The rescue personnel quickly locate the fault through holographic display and voice guidance and cooperates with the experts in real time to dispose. Step S7.4, the system automatically generates a rescue report, records the fault type, rescue path and time-consuming data, and iteratively optimizes the prediction model and rescue guide algorithm based on rescue feedback.

8. A fault remote diagnosis system for fusing switch quantity monitoring data and multi-source service data, for implementing the fault remote diagnosis method for fusing switch quantity monitoring data and multi-source service data according to any one of claims 1-7, characterized in that, It includes a sensing network, an edge computing node, a remote monitoring platform, an MR terminal and a mobile terminal. The sensing network is electrically connected with the device controller, used for collecting device on-off quantity monitoring data, environmental data and running state data. The on-off quantity monitoring data includes 13 types of core on-off quantity data, and the data is transmitted through an encryption protocol. The edge computing node is equipped with an RTOS operating system, communicates with the sensing network and the remote monitoring platform through a qNCR+5G communication network, and performs data cleaning, filtering, feature extraction and AI reasoning tasks, supporting dynamic switching of high-precision and lightweight AI models. The remote monitoring platform has a JSON format database with the device emergency disposal platform device number as the unique key, integrates a digital twin modeling module, a dual-source diagnosis module and a result fusion module. The digital twin modeling module constructs a multi-dimensional dynamic simulation model combined with device basic parameters and real-time data. The dual-source diagnosis module includes a business data fault prediction model based on attention mechanism-Bi-LSTM and a on-off quantity expert system containing a dynamic updating knowledge base. The result fusion module uses an improved D-S evidence theory to process the conflict of dual-source diagnosis results. The MR terminal is wirelessly connected with the remote monitoring platform, realizes virtual-real registration through the fusion of QR code manual identification and SIFT natural feature points, provides 3D holographic fault display and expert remote marking guidance. The mobile terminal is used to issue post-rescue verification instructions to form a closed loop of fault diagnosis, rescue and verification.

9. A computer device, comprising: The computer program stored in the memory and executable on the processor, when the processor executes the computer program, realizes the fault remote diagnosis method of fusing device on-off quantity monitoring data and multi-source business data according to any one of claims 1-7.

Images