A work platform fault diagnosis system and method based on multi-sensor data fusion

Abstract

The application discloses a kind of based on multi-sensing data fusion's job platform fault diagnosis system and method, system includes data acquisition module, data communication module, man-machine interaction module, image recognition module and intelligent fault diagnosis module;The process for fault diagnosis of the present application is as follows: analyze the mechanical structure of job platform, comb job platform fault;Establish job platform state data acquisition model;Based on job platform data acquisition model, establish feature data set, use Gaussian distribution to construct the Gaussian membership function distribution model of each fault feature, establish binary group BPA uncertain information model;Based on the weighted average method of evidence correlation coefficient, the obtained BPA model is fused, the fusion result is compared with the set threshold Δ, and the final diagnostic result is obtained.The present application solves the technical problems that the current job platform fault diagnosis application is less, the real-time monitoring data volume is large, the overhaul and maintenance workload is large, the diagnostic efficiency is low and the degree of automation is low.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis for railway maintenance work platforms, and in particular to a fault diagnosis system and method for work platforms based on multi-sensor data fusion. Background Technology

[0002] Currently, the dedicated maintenance equipment and facilities for high-speed trains within the railway system are still mainly based on planned preventive maintenance. With the development of science and technology, the equipment and facilities are being digitally empowered and intelligently upgraded. The inherent operational data value of the equipment and facilities is being fully explored to monitor, evaluate, and diagnose the health status of the equipment and facilities. The exploration of building a more scientific, precise, and economical maintenance management system is underway to reduce the maintenance costs of the equipment and facilities and extend the service life of the equipment. The conditions for construction and application are gradually being met.

[0003] The three-tiered work platform and inspection warehouse columns have the same longitudinal span, consisting of modules several meters long. Each module is a frame structure composed of two H-shaped steel longitudinal beams and other steel crossbeams, with patterned steel plates paving the treads. When climbing to the top is required, the flip-up device on one side of the platform and the tilting frame device with the suspended safety net on the other side operate simultaneously under the push of cylinders, forming a working surface to support the workers on the roof for walking and inspection work.

[0004] During the movement of the flipping plate and tilting frame, various malfunctions are inevitable, posing serious safety hazards. Therefore, it is necessary to promptly inspect the status of the work platform to ensure safe operation.

[0005] Existing technologies primarily employ expert system-based fault diagnosis methods. Currently, the third-level work platform in the EMU maintenance workshop is still mainly inspected manually at regular intervals, with workers periodically monitoring and evaluating its status. Dispatch room staff use spherical cameras at both ends of the workshop for remote control, periodically observing the movement of the flaps along the entire line as a secondary measure. While this method has some application value, it struggles to address complex and ever-changing fault diagnosis needs in a timely manner and poses certain safety hazards.

[0006] Currently, existing technologies have the following main shortcomings:

[0007] (1) When manually inspecting the three-layer platform working surface system, it is necessary to check each cylinder and tilting frame. This inspection method will consume a lot of manpower and time to fully determine the safety status of the working platform, resulting in an increase in labor costs.

[0008] (2) As the equipment is used for a longer period of time, cylinders, flaps and other equipment will age. When a fault occurs, traditional inspection methods cannot detect it in time and make a quick response, which can easily lead to safety accidents and cause personal injury.

[0009] (3) The traditional maintenance method has too few inspection indicators.

[0010] Therefore, how to efficiently and accurately complete the fault diagnosis of the work platform in real time has become an urgent problem to be solved. Summary of the Invention

[0011] Purpose of the invention: To address the shortcomings of existing technologies, this invention proposes a fault diagnosis system and method for work platforms based on multi-sensor data fusion, which solves the technical problems of limited application of fault diagnosis for work platforms, large amount of real-time monitoring data, large workload of maintenance, low diagnostic efficiency, and low degree of automation.

[0012] Technical solution: The fault diagnosis system for work platforms based on multi-sensor data fusion of the present invention includes a data acquisition module, a data communication module, a human-computer interaction module, an image recognition module, and an intelligent fault diagnosis module;

[0013] The data acquisition module acquires real-time air pressure data, voltage data, current data, and the actual operating status of the flapper on the work platform;

[0014] The data communication module, including a serial port server and a switch, is used to transmit the values ​​from the analog signal acquisition card and PLC to the human-machine interface module in real time via the TCP / IP protocol.

[0015] The human-computer interaction module receives and parses the data acquisition model in real time, displays the status and manages fault information on the PC, preprocesses fault information, and can also be displayed on mobile terminals, making it more convenient to use and more diversified in human-computer interaction.

[0016] The image recognition module is used to acquire and process images of the flip-up status, and the proximity switch is used to make multi-angle judgments on the actual operating status of the flip-up.

[0017] The intelligent fault diagnosis module is used to perform intelligent fault diagnosis based on data acquired by the data acquisition module, data communication module, human-computer interaction module, and image recognition module, to determine the fault location and fault type, and to propose fault handling suggestions.

[0018] The fault diagnosis method for work platforms based on multi-sensor data fusion of this invention is implemented using a fault diagnosis system for work platforms based on multi-sensor data fusion. The diagnosis method includes the following steps:

[0019] (1) Analyze the mechanical structure of the work platform and identify its faults;

[0020] (2) Establish a work platform status data acquisition model, which includes a solenoid valve status acquisition model, a flapper status acquisition model and a cylinder status acquisition model.

[0021] (3) Based on the data acquisition model of the work platform, a feature dataset is established, and a Gaussian membership function distribution model of each fault feature is constructed using Gaussian distribution. A binary BPA uncertain information model is established.

[0022] (4) The BPA model obtained by the weighted average method based on the evidence correlation coefficient is fused, and the fusion result is compared with the set threshold Δ to obtain the final diagnosis result.

[0023] In step (2), when establishing the flip-board status acquisition model, the proximity switch value is read by PLC and the status of the flip-board is identified by visual recognition method to determine the flip-board movement status.

[0024] In step (2), when using the visual recognition method to identify the status of the flip panel, the PLC reads the proximity switch value to identify the status of the flip panel, obtains the identification image, establishes the image training set, and performs image preprocessing; the established identification target feature model is matched with the obtained training set.

[0025] The process of step (3) is as follows:

[0026] Step (3.1): Select training samples and test samples, and set the fault categories;

[0027] Step (3.2): Construct Gaussian models for current, voltage, and air pressure attributes; calculate the mean and standard deviation of training samples for various fault categories under current, voltage, and air pressure.

[0028] Step (3.3) involves matching the Gaussian model to obtain the BPA.

[0029] In step (3.2), the Gaussian models for the current, voltage, and air pressure properties are constructed as follows:

[0030]

[0031]

[0032]

[0033] Where, μ C (x) is the Gaussian model of class F1 on attribute current C; μ V (x) is the Gaussian model of category F1 on attribute voltage V; μ P (x) is the Gaussian model of category F1 on attribute air pressure P.

[0034] In step (3.2), the average value under the attribute current C is and the sample standard deviation is σ C The average value under the attribute voltage V is and the sample standard deviation is σ V The average value under attribute P is and the sample standard deviation is σ P ;Right now

[0035]

[0036]

[0037]

[0038] in, This represents the value of the l-th training sample of class F1 on attribute C; This represents the value of the l-th training sample of class F1 on attribute V; This represents the value of the l-th training sample of category F1 on attribute P.

[0039] Step (4) includes the following process:

[0040] Step (4.1), integrating the BPA model, the process is as follows:

[0041] Step (4.1.1) calculates the evidence correlation coefficient to obtain the similarity matrix SM:

[0042]

[0043] Among them, the evidence correlation coefficient S between m1 and m2 12 The correlation coefficient of evidence between m1 and m3 is S. 13 The correlation coefficient of evidence between m2 and m3 is S. 23 ;

[0044] Step (4.1.2) calculates the discount factor for each BPA and weights them to obtain the average evidence m. A First, calculate the confidence level Crd(m). i )

[0045]

[0046] Step (4.1.3) yields the fusion result from the Dempster combination fusion.

[0047] Step (4.2) involves fusing the results and making a decision to obtain the final outcome.

[0048] In step (4.1.2), the obtained confidence level Crd(m) is used. i (i = 1, 2, 3) We calculate the weighted average evidence m by weighting each corresponding BPA. A :

[0049]

[0050] In step (4.2), during fault diagnosis, a threshold Δ is set. When the confidence level of a fault reaches the threshold Δ, the platform is considered to have malfunctioned. The fused m... F Compared with Δ, when m F When the value is greater than Δ, the platform is considered to have malfunctioned.

[0051] Working principle: The fault diagnosis system for work platforms based on multi-sensor data fusion of the present invention can quickly determine the fault location and fault type, and put forward fault handling suggestions to ensure safe production operation.

[0052] This invention presents a fault diagnosis method for work platforms based on multi-sensor data fusion. By studying the mechanical structure and control principles of a three-layer work platform, and identifying fault phenomena and patterns, a work platform status data acquisition model is designed based on the existing three-layer work platform. Different types of sensors are used to acquire the detection values ​​of each unit at each time point as test samples. A Gaussian distribution model is used to construct a membership degree distribution model for training samples across various attributes. The test samples are matched with the membership degree distribution model to establish a binary pair uncertainty information model (BPA). The obtained BPA model is then fused using a weighted average method based on the evidence correlation coefficient. The fused decision result is compared with a set threshold Δ to obtain the final diagnostic result.

[0053] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0054] (1) This invention establishes a data acquisition model and uses multiple indicators to achieve a comprehensive evaluation and analysis of the entire operating platform, establishes a fault feature dataset, establishes a binary BPA uncertain information model, and the fault types obtained after fusing BPA are more accurate, thereby improving the accuracy and efficiency of the diagnostic results.

[0055] (2) The present invention preprocesses the fault signals and fault types of the operating platform during operation through the human-computer interaction system, and uses the intelligent fault diagnosis module to accurately and quickly locate the fault location, remind the maintenance personnel in real time, and give handling suggestions. While improving maintenance efficiency, it also ensures maintenance quality, so that the platform can quickly return to normal operation after a fault, and ensures the safe and stable operation of the operating platform. Attached Figure Description

[0056] Figure 1 This is a flowchart of the fault diagnosis method for an operating platform based on multi-sensor data fusion according to the present invention;

[0057] Figure 2 The flowchart for step (2) of the fault diagnosis method for an operating platform based on multi-sensor data fusion is shown below.

[0058] Figure 3 The flowchart for step (3) of the fault diagnosis method for an operating platform based on multi-sensor data fusion is shown below.

[0059] Figure 4 The flowchart for step (4) of the fault diagnosis method for an operating platform based on multi-sensor data fusion is shown below.

[0060] Figure 5 This is a topology diagram of a fault diagnosis system for an operating platform based on multi-sensor data fusion. Detailed Implementation

[0061] Example:

[0062] The present invention provides a fault diagnosis system for a work platform based on multi-sensor data fusion, comprising a data acquisition module, a data communication module, a human-computer interaction module, an image recognition module, and an intelligent fault diagnosis module.

[0063] The data acquisition module is used to acquire real-time air pressure data, voltage data, current data, and the actual operating status of the flapper on the work platform.

[0064] The data communication module includes a serial port server and a switch, which are used to transmit the values ​​from the analog signal acquisition card and PLC to the human-machine interface module in real time via the TCP / IP protocol.

[0065] The human-computer interaction module receives and parses the data acquisition model in real time, displays the status and manages fault information on the PC, and preprocesses the fault information. It can also be displayed on mobile terminals, making it more convenient to use and more diversified in human-computer interaction.

[0066] The image recognition module is used to acquire and process images of the flip-up panel's status, and the proximity switch is used to make multi-angle judgments on the actual operating status of the flip-up panel.

[0067] The intelligent fault diagnosis module uses data acquired by the data acquisition module, data communication module, human-computer interaction module, and image recognition module to perform intelligent fault diagnosis, determine the fault location and fault type, and propose fault handling suggestions.

[0068] like Figure 1 As shown, the fault diagnosis method for an operating platform based on multi-sensor data fusion of the present invention includes the following steps:

[0069] Step (1): Analyze the mechanical structure of the work platform and identify the fault phenomena of the work platform;

[0070] In this embodiment, the work platform is a fully suspended type, with the platform suspension frame and the lower chord of the building frame connected by bolts. The platform is 450 meters long and is divided into a working surface system and a safety net frame system. One side of the working surface system has 96 flaps, and the other side of the safety net frame system has 96 tilting frames. Each flap and tilting frame is composed of a frame structure made of H-beam longitudinal beams, with patterned steel plates laid on the tread. The cylinders are controlled by solenoid valves and throttle valves to perform 90° retraction and descent actions. Work platform malfunctions include flaps not rising fully, solenoid valve short circuits, cylinders lacking power, and excessive air pressure in the air circuit. Failure to raise the flap fully is due to debris in the cylinder preventing the piston from moving to the predetermined stroke length, resulting in the flap not being in a vertical position when retracted. A solenoid valve short circuit is caused by leaking seals or a burnt-out coil, preventing the solenoid valve from being energized and causing a short circuit. Lack of cylinder power is due to aging cylinder gaskets, causing a decrease in the seal between the cylinder and cylinder head, affecting the internal pressure of the cylinder and resulting in insufficient cylinder power.

[0071] Step (2) Establish the status data acquisition model of the work platform. The status data acquisition model of the work platform includes the status acquisition model of the solenoid valve, the status acquisition model of the flapper and the status acquisition model of the cylinder.

[0072] Step (2.1) establishes the solenoid valve status acquisition model. The solenoid valve status acquisition model utilizes 48 AC current transmitters and 4 AC voltage transmitters to acquire the current and voltage data of the solenoid valve. The 48 AC current transmitters are divided into 24 rising current transmitters and 24 falling current transmitters. Each group of flap units and fence units at the same position on both sides of the track shares the same rising and falling current. Each group of 6 flap units and fence units at the same position on both sides of the track shares the same voltage value. The current and voltage data are analog quantities of 4-20mA, acquired via an analog data acquisition card using the RS485 communication protocol.

[0073] Step (2.2) establishes the flapper status acquisition model. The flapper status acquisition model includes PLC reading proximity switch values ​​and using visual recognition technology to identify the flapper's status and determine its motion state; the process is as follows:

[0074] In step (2.2.1), the PLC reads the proximity switch values ​​to identify the status of the flaps. There are 96 flaps on one side of the working face system and 96 tilting frames on one side of the safety net frame system. Each flap and tilting frame is equipped with a proximity switch, which has a boolean value of 0 or 1, obtained through the PLC. The status changes from 1 to 0 when descending from a vertical position and from 0 to 1 when rising from a horizontal position. In the PLC, a signal of 1 indicates that the flap is in a vertically retracted state, and a signal of 0 indicates that the flap is in any other state.

[0075] Step (2.2.2) employs a visual recognition method to identify the status of the flip-up panels. Twenty-four sets of cameras are installed at a 45° downward angle on the platform supports. Each camera reads images from eight flip-up panels, and each hard disk recorder saves video images from six cameras using the TCP / IP protocol. Yellow stickers (50cm x 10cm rectangles) are affixed to the side of each flip-up panel in both the working surface system and the safety mesh system near the track as identification targets. Halcon, a machine vision library, is used for image processing and analysis in industrial and scientific fields. In Halcon, the area of ​​the feature region in the training set is obtained through operators. In practical applications, real-time images are matched with the training set to determine the movement status of the flip-up panels.

[0076] Step (2.2.2.1) involves acquiring the recognition images and establishing an image training set. Camera images are captured in real-time using screenshot code at 5-second intervals. A training set of 24 sets is established by capturing video images from different scenes: daytime, cloudy days, and nighttime.

[0077] Step (2.2.2.2) involves image preprocessing. Image preprocessing includes grayscale conversion, filtering, and contrast enhancement. Since the dominant color of the actual image acquired on-site is white, and the identified target is yellow with a significant difference from the surrounding colors, grayscale conversion is used. Specifically, thresholding is applied to select targets with grayscale values ​​between 200 and 250 to separate the target recognition region for the first step. From the separated target region, targets with widths between 400 and 99999 are selected for width filtering. Then, targets with heights between 150 and 99999 are selected from the filtered targets for height filtering to obtain the target recognition region for the second step. Finally, targets with areas between 150 and 99999 are selected for area filtering to extract the final target feature model.

[0078] Step (2.2.2.3) involves matching the target feature model. The FindShapeModel function in Halcon is used to match the established target feature model with the obtained training set.

[0079] Step (2.3) establishes the cylinder status acquisition model. In this embodiment, the cylinder status acquisition model includes 48 pressure sensors to monitor the actual pressure values ​​of the air circuits of the 24 flap modules of the working face system and the 24 fence modules of the safety net system. The 48 pressure sensors are evenly deployed in the air circuits, and timely and effective early warnings are issued when insufficient pressure occurs to protect the safety of on-site personnel. Among them, the air pressure values ​​are analog data of 4-20mA, and analog quantities such as air pressure are acquired through an analog data acquisition card using the RS485 communication protocol. When climbing to the top for inspection, first confirm whether the pressure of the pneumatic system is greater than 0.5MPa, manually operate the two flaps to lower smoothly, and when the maintenance is completed, manually press the button to retract all flaps. After ten seconds, the flaps return to the vertical position and the button is released. If the air pressure is insufficient, the flaps are manually restored to the vertical position.

[0080] Step (3): Based on the data acquisition model of the work platform, establish a feature dataset, use Gaussian distribution to construct a Gaussian membership function distribution model for each fault feature, and establish a binary BPA uncertain information model.

[0081] Step (3.1) involves selecting training and testing samples. Three fault categories are set: F1 = flapper not fully raised, F2 = solenoid valve short circuit, and F3 = cylinder lack of power, forming the identification framework {F1, F2, F3}. Each type has three features: current C, voltage V, and air pressure P. For the original dataset, 30 samples are continuously collected for 30 seconds as training samples, and 30 samples are continuously collected at 5-minute intervals as testing samples. The following will use category F1 as an example; categories F2 and F3 are calculated similarly.

[0082] Step (3.2) involves constructing Gaussian models for each attribute;

[0083] Step (3.2.1) involves calculating the mean and sample standard deviation. Specifically, it involves calculating the mean and sample standard deviation of the 30 training samples corresponding to category F1 under the attributes of current C, voltage V, and air pressure P. The mean under attribute current C is... and the sample standard deviation is σ C The average value under the attribute voltage V is and the sample standard deviation is σ V The average value under attribute P is and the sample standard deviation is σ P .Right now

[0084]

[0085]

[0086]

[0087] in, This represents the value of the l-th training sample of class F1 on attribute C; This represents the value of the l-th training sample of class F1 on attribute V; This represents the value of the l-th training sample of category F1 on attribute P.

[0088] Step (3.2.2) involves constructing a Gaussian model based on the calculation results. Specifically, based on the obtained mean and standard deviation, a Gaussian model of category F1 is constructed under the attribute current C, voltage V, and air pressure P.

[0089]

[0090]

[0091]

[0092] Where, μ c (x) is the Gaussian model of class F1 on attribute C; μ V (x) is the Gaussian model of category F1 on attribute V; μ P (x) is the Gaussian model of category F1 on attribute P.

[0093] Step (3.3): Match the Gaussian model to obtain the BPA. Randomly select a sample from category F1 in the test samples as the test sample, match it with the corresponding Gaussian model under each attribute, obtain the matching degree between the test sample and category F1, and generate the BPA model of the test sample.

[0094] H(Ft)=μ F (x)| x＝t

[0095] H(F←t) reflects the degree of matching between the test sample and F, and is determined by the intersection points between the Gaussian model samples corresponding to F. From this formula, we obtain the BPA model of the test sample under attributes C, V, and P: m1(F1), m1(F1, F2), m1(F2, F3), m1(F1, F2, F3) are the BPA of the test sample on attribute C; m2(F1), m2(F1, F2), m2(F2, F3), m2(F1, F2, F3) are the BPA of the test sample on attribute V; and m3(F1), m3(F1, F2), m3(F2, F3), m3(F1, F2, F3) are the BPA of the test sample on attribute P.

[0096] Step (4): The BPA model obtained is fused using the weighted average method based on the evidence correlation coefficient. The fusion result is compared with the set threshold Δ to obtain the final diagnosis result.

[0097] Step (4.1), integrate the BPA model;

[0098] Step (4.1.1) calculates the evidence correlation coefficient to obtain the similarity matrix. Since the calculated BPA may be 0, indicating a significant conflict, conflict resolution is performed before fusion. Evidence correlation coefficient r BPA The correlation coefficient measures the degree of interrelation between pieces of evidence. A higher correlation coefficient indicates a stronger correlation and less conflict between the pieces of evidence. The calculated correlation coefficient r... BPA As the support Sup(m) i m j ), abbreviated as S ij That is, Sup(m) i m j ) = r BPA (m i m j ).

[0099]

[0100] The correlation coefficient S between m1 and m2 is calculated using the formula. 12 The correlation coefficient of evidence between m1 and m3 is S. 13 The correlation coefficient of evidence between m2 and m3 is S. 23 This yields a similarity matrix SM:

[0101]

[0102] Where m1 is the BPA of F1 on attribute C, m2 is the BPA of F2 on attribute V, and m3 is the BPA of F3 on attribute P.

[0103] Step (4.1.2) calculates the discount factor for each BPA and weights them to obtain the average evidence m. A First, calculate the confidence level Crd(m). i )

[0104]

[0105] The credibility of the evidence, Crd(m), is calculated from the calculated Sup(m1), Sup(m2), and Sup(m3). i (i = 1, 2, 3) i.e., Crd(m1), Crd(m2), Crd(m3), using the obtained confidence level Crd(m i (i = 1, 2, 3) We calculate the weighted average evidence m by weighting each corresponding BPA. A :

[0106]

[0107] That is, m A (F1), m A (F2), m A (F3), m A (F1, F2), m A (F1, F3), m A (F2, F3), m A (F1, F2, F3).

[0108] Where, m A (F1) represents the average evidence after F1 weighting, m A (F2) represents the average evidence after F2 weighting, m A (F3) represents the average evidence after F3 weighting, m A (F1, F2) represents the weighted average evidence of F1 and F2, m A (F1, F3) represents the weighted average evidence of F1 and F3, m A (F2, F3) represents the weighted average evidence of F2 and F3, m A (F1, F2, F3) is the weighted average of F1, F2, and F3.

[0109] Step (4.1.3) involves obtaining the fusion result m from the Dempster combination fusion. F Dempster's combinatorial rules are a mathematical method for probabilistic reasoning and decision-making under conditions of uncertainty. Let m... A Using Dempster's combinatorial rule to fuse itself 3 times, we get m. F ,Right now

[0110]

[0111] Get m F (F1), m F (F2), m F (F3), m F (F1, F2), m F (F1, F3), m F (F2, F3), m F (F1, F2, F3), m F (F1) is the final evidence after F1's own fusion, m F (F2) is the final evidence after F2's own fusion, m F (F3) is the final evidence after F3's own fusion, m F (F1, F2) represents the final evidence after the fusion of F1 and F2, m F (F1, F3) represents the final evidence after the fusion of F1 and F3 themselves, mF (F2, F3) represents the final evidence after the fusion of F2 and F3 themselves, m F (F1, F2, F3) represents the final evidence after F1, F2, and F3 are integrated.

[0112] Step (4.2) involves fusing the results to arrive at a decision. During fault diagnosis, a threshold Δ is set; when the confidence level of a fault reaches this threshold Δ, the platform is considered to have malfunctioned. The fused m... F Compared with Δ, when m F When the value is greater than Δ, it is assumed that the platform has one of the faults in {F1, F2, F3}.

Claims

1. A fault diagnosis system for an operating platform based on multi-sensor data fusion, characterized in that: It includes a data acquisition module, a data communication module, a human-computer interaction module, an image recognition module, and an intelligent fault diagnosis module; The data acquisition module acquires real-time air pressure data, voltage data, current data, and the actual operating status of the flapper on the work platform. The data communication module includes a serial port server and a switch, which are used to transmit the values ​​of the analog signal acquisition card and PLC to the human-machine interaction module in real time via TCP / IP protocol; The human-computer interaction module receives and parses the data acquisition model in real time, displays the status and manages fault information on the PC, preprocesses fault information, and can also be displayed on mobile terminals, making it more convenient to use and more diversified in human-computer interaction. The image recognition module is used to acquire and process images of the flip-board status, and the proximity switch is used to make multi-angle judgments on the actual operating status of the flip-board. The intelligent fault diagnosis module is used to perform intelligent fault diagnosis based on data acquired by the data acquisition module, data communication module, human-computer interaction module and image recognition module, to determine the fault location and fault type, and to propose fault handling suggestions.

2. A fault diagnosis method for an operating platform based on multi-sensor data fusion, characterized in that: The fault diagnosis system for an operating platform based on multi-sensor data fusion as described in claim 1 is used, and the diagnosis method includes the following steps: (1) Analyze the mechanical structure of the work platform and identify its faults; (2) Establish a work platform status data acquisition model, which includes a solenoid valve status acquisition model, a flapper status acquisition model and a cylinder status acquisition model. (3) Based on the data acquisition model of the work platform, a feature dataset is established, and a Gaussian membership function distribution model of each fault feature is constructed using Gaussian distribution. A binary BPA uncertain information model is established. (4) The BPA model obtained by the weighted average method based on the evidence correlation coefficient is fused, and the fusion result is compared with the set threshold Δ to obtain the final diagnosis result.

3. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 2, characterized in that: In step (2), when establishing the flip-board status acquisition model, the proximity switch value is read by PLC and the status of the flip-board is identified by visual recognition method to determine the flip-board movement status.

4. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 3, characterized in that: In step (2), when using the visual recognition method to identify the status of the flip panel, the PLC reads the proximity switch value to identify the status of the flip panel, obtains the identification image, establishes the image training set, and performs image preprocessing; the established identification target feature model is matched with the obtained training set.

5. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 2, characterized in that: The process of step (3) is as follows: Step (3.1): Select training samples and test samples, and set the fault categories; Step (3.2) involves constructing Gaussian models for the current, voltage, and air pressure properties; Calculate the mean and standard deviation of the training samples corresponding to various fault categories under current, voltage, and air pressure. Step (3.3) involves matching the Gaussian model to obtain the BPA.

6. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 5, characterized in that: In step (3.2), the Gaussian models for the current, voltage, and air pressure properties are constructed as follows: Where, μ C (x) is the Gaussian model of class F1 on attribute current C; μ V (x) is the Gaussian model of category F1 on attribute voltage V; μ P (x) is the Gaussian model of category F1 on attribute air pressure P.

7. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 6, characterized in that: In step (3.2), the average value under the attribute current C is and the sample standard deviation is σ C The average value under the attribute voltage V is and the sample standard deviation is σ V The average value under attribute P is and the sample standard deviation is σ P ;Right now in, This represents the value of the k-th training sample of class F1 on attribute C; This represents the value of the k-th training sample of class F1 on attribute V; This represents the value of the l-th training sample of category F1 on attribute P.

8. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 2, characterized in that: Step (4) includes the following process: Step (4.1), integrating the BPA model, the process is as follows: Step (4.1.1) calculates the evidence correlation coefficient to obtain the similarity matrix SM: Among them, the evidence correlation coefficient S between m1 and m2 12 The correlation coefficient of evidence between m1 and m3 is S. 13 The correlation coefficient of evidence between m2 and m3 is S. 23 ; Step (4.1.2) calculates the discount factor for each BPA and weights them to obtain the average evidence m. A First, calculate the confidence level Crd(m). i ) Step (4.1.3) yields the fusion result from the Dempster combination fusion. Step (4.2) involves fusing the results and making a decision to obtain the final outcome.

9. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 8, characterized in that: In step (4.1.2), the obtained confidence level Crd(m) is used. i (i = 1, 2, 3) We calculate the weighted average evidence m by weighting each corresponding BPA. A :

10. The fault diagnosis method for an operating platform based on multi-sensor data fusion according to claim 8, characterized in that: In step (4.2), during fault diagnosis, a threshold Δ is set. When the confidence level of a fault reaches the threshold Δ, the platform is considered to have malfunctioned. The fused m... F Compared with Δ, when m F When the value is greater than Δ, the platform is considered to have malfunctioned.

Images