A fault detection method, system and device for a smart water station

Abstract

This application relates to the field of equipment fault detection technology, and in particular to a fault detection method, system, and device for a smart water station. The method proposed in this application is used in a device including a pontoon body, guide rails, a lifting module, and a vision acquisition component. It uses a position detection component and a first image sensor to jointly determine whether an abnormal response exists. If the conditions for entering fault identification are met, the target structural area of ​​the pontoon is determined based on the response interruption point. A second image sensor is then controlled to acquire the target image, and the fault type of the pontoon is identified through structural feature extraction and a fault identification network. This application can achieve precise coordination between fault location and identification, improving detection efficiency and identification accuracy.

Description

Technical Field

[0001] This application relates to the field of equipment fault detection technology, and in particular to a fault detection method, system and device for a smart water station. Background Technology

[0002] In existing smart water station structures, to adapt to dynamic water level changes and ensure the continuous operation of water intake equipment, a float-type lifting structure is typically used to achieve real-time adjustment of the water intake depth. During the lifting and lowering of the float within the guide rail, its displacement is usually monitored by position detection components (such as encoders). However, under long-term operation or in complex water environments, the float may experience malfunctions such as jamming, structural deformation, or guide rail misalignment. Traditional displacement detection methods cannot accurately identify the specific location and type of these malfunctions.

[0003] Problems with existing technologies: Current float monitoring methods are mostly based on fixed angles and continuous monitoring, resulting in simple identification logic, high computational load, and difficulty in achieving efficient and accurate fault identification. To solve these problems, this application designs a fault detection method, system, and device for smart water stations. Summary of the Invention

[0004] The technical problem to be solved by this application is to address the shortcomings of the prior art by providing a fault detection method, system and device for a smart water station. The method proposed in this application is used in a device including a pontoon body, guide rail, lifting module and vision acquisition component. The position detection component and the first image sensor jointly determine whether there is an abnormal response. If the conditions for entering fault identification are met, the target structural area of ​​the pontoon is determined based on the response interruption point, the second image sensor is controlled to acquire the target image, and the fault type of the pontoon is identified through structural feature extraction and fault identification network.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A fault detection method for a smart water station is applied to a fault detection device. The fault detection device includes a lifting module, a visual acquisition component, a guide rail, and a float body. The lifting module is used to raise and lower the float body in the water in response to the water level of the water station to be detected. The visual acquisition component includes a first image sensor and a second image sensor. The first image sensor is disposed in the guide rail at a position perpendicular to the float body, and the second image sensor is disposed on the outside of the float body. The method includes:

[0007] Based on the acquired water level information, the lifting module is controlled to drive the float body to rise and fall along the guide rail;

[0008] The system determines whether the conditions for entering the fault identification stage are met based on the position detection component in the lifting module and the first image sensor.

[0009] If satisfied, the second image sensor executes the float fault identification logic to obtain the float fault type. The float fault identification logic locates the target area of ​​the float body based on the response interruption point and performs structural feature extraction and fault type identification processing on the target area.

[0010] The pontoon fault identification logic includes:

[0011] Calculate the response interruption point based on the feedback information of the lifting module during the current action cycle and the detection information of the first image sensor;

[0012] Based on the response interruption point, the target area of ​​the pontoon body is determined, and a target image of the target area is obtained through the second image sensor;

[0013] The target image is subjected to image enhancement, edge detection, and contour segmentation to obtain structural features;

[0014] The structural features are input into a preset fault identification network, which processes the structural features and outputs the buoy fault type. The fault identification network is trained using historical fault images of the buoy.

[0015] The step of calculating the response interruption point based on the feedback information within the current operation cycle of the lifting module and the detection information from the first image sensor includes:

[0016] Obtain the target displacement command and displacement feedback information of the lifting module in the current action cycle, and construct the theoretical motion trajectory of the float body in the current action cycle;

[0017] Based on the viewing angle parameters and calibration parameters of the first image sensor, the theoretical motion trajectory is back-projected in the image space to obtain the theoretical change area at the top of the pontoon.

[0018] The theoretical change region is compared with the actual change region of the top of the pontoon in the detection information using optical flow method to detect the change in response trend between consecutive image frames, so as to obtain the start time of the interruption of the image response at the top of the pontoon, and the control segment corresponding to the start time is taken as the response interruption point.

[0019] Based on the response interruption point, the target area of ​​the pontoon body is determined, including:

[0020] The movement path of the pontoon body along the guide rail is divided into multiple preset structural sections, wherein each preset structural section corresponds one-to-one with the structural area of ​​the pontoon body.

[0021] The corresponding preset structural segment is determined based on the control segment of the response interruption point;

[0022] The structural area of ​​the corresponding pontoon body in the corresponding preset structural section is taken as the target area.

[0023] The fault identification network includes a structure encoding layer, a state reasoning layer, and a fault classification output layer, wherein:

[0024] The structural coding layer is used to vectorize the input structural features, including edge continuity parameters, contour closure index, and occlusion area ratio.

[0025] The state reasoning layer is equipped with a graph structure attention mechanism, which is used to establish spatial correlations between structure encoding vectors and reason about the integrity state and stress consistency characteristics of the pontoon structure.

[0026] The fault classification output layer is used to output the fault type of the buoy through the Softmax classifier based on the integrity status and force consistency characteristics. The fault types include buoy jamming, guide rail jamming, bottom floating obstruction, and buoy attitude tilting.

[0027] The lifting module includes a water level sensor, a position detection component, and a lifting mechanism. Controlling the lifting module to drive the float body to rise and fall along the guide rail based on the acquired water level information includes:

[0028] Obtain the water level information collected by the water level sensor;

[0029] The water level information is compared with the preset target water intake depth to determine the target position of the pontoon body;

[0030] The lifting mechanism is controlled to move the pontoon body to the height corresponding to the target position.

[0031] The system determines whether the conditions for entering the fault identification stage are met based on the position detection component in the lifting module and the first image sensor, including:

[0032] Obtain the first displacement data output by the position detection component;

[0033] The top image sequence acquired by the first image sensor is obtained, wherein the acquisition period of the first image sensor is set synchronously with the action period of the lifting module;

[0034] Edge detection is performed sequentially on the top images in the top image sequence to obtain a contour feature sequence;

[0035] According to a pre-defined spatial height mapping model, the pixel size changes of the contour feature sequence in the image are converted into vertical displacement values ​​to obtain second displacement data, wherein the spatial height mapping model is constructed based on the image projection transformation relationship.

[0036] The conditions for entering the fault identification stage include at least one of the following:

[0037] The absolute value of the difference between the first displacement data and the second displacement data is greater than or equal to a preset error threshold.

[0038] The second displacement data did not change during the operation cycle;

[0039] The direction of change of the second displacement data is opposite to the direction of change of the first displacement data.

[0040] A fault detection system for a smart water station is applied to a fault detection device. The fault detection device includes a lifting module, a vision acquisition component, a guide rail, and a float body. The lifting module is used to raise and lower the float body in the water in response to the water level of the water station to be detected. The vision acquisition component includes a first image sensor and a second image sensor. The first image sensor is disposed in the guide rail at a position perpendicular to the float body, and the second image sensor is disposed on the outside of the float body. The system includes:

[0041] The water level control module is used to acquire the water level information of the water station to be tested, and control the lifting module to move the float body to the height corresponding to the target position according to the water level information;

[0042] The displacement determination module is used to determine whether the conditions for entering the fault identification stage are met based on the first displacement data output by the position detection component in the lifting module and the top image sequence collected by the first image sensor.

[0043] The image acquisition module is used to control the second image sensor to acquire target images of the target area of ​​the float body after the conditions for entering the fault identification stage are met.

[0044] The feature extraction module is used to perform image enhancement, edge detection, and contour segmentation on the target image to obtain the structural features of the pontoon;

[0045] The fault identification module is used to input the structural features into a preset fault identification network and output the corresponding float fault type based on the fault identification network.

[0046] A fault detection device for a smart water station, the fault detection device being used to detect the type of fault in the float body, wherein the fault detection device includes:

[0047] Guide rails are used to limit the vertical movement of the pontoon body along its length;

[0048] The lifting module is used to drive the float body in the water to rise and fall along the guide rail in response to the water level of the water station to be tested.

[0049] The visual acquisition component includes a first image sensor and a second image sensor. The first image sensor is disposed in the guide rail at a position perpendicular to the pontoon body, and the second image sensor is disposed on the outside of the pontoon body.

[0050] A controller, connected to the lifting module and the vision acquisition component, is configured as follows:

[0051] Obtain water level information and control the raising and lowering of the buoy;

[0052] Based on the displacement feedback of the lifting module and the top image sequence of the first image sensor, it is determined whether the conditions for entering the fault identification stage are met.

[0053] When the conditions for entering the fault identification stage are met, the buoy response interruption point is calculated, the target area of ​​the buoy body is determined, and the second image sensor is controlled to acquire the target image.

[0054] Structural features are extracted from the target image and input into a preset fault identification network to output the fault type of the pontoon.

[0055] Compared with the prior art, the beneficial effects of this application are:

[0056] This application achieves dual-channel monitoring of the pontoon's operating status by combining displacement feedback information from the lifting module with top image sequences acquired by the first image sensor. It also introduces a computational mechanism to respond to interruptions, enabling accurate location of the time period and corresponding structural area where a pontoon malfunctions during operation. Based on this, a second image sensor is controlled to precisely acquire data from the target area. Combined with structural feature extraction and a trained fault identification network, the specific fault type of the pontoon is effectively identified. Compared to existing solutions relying on a single sensor or passive monitoring, this method significantly improves the accuracy of fault location, the precision of identification, and overall processing efficiency, making it widely applicable to complex smart water station scenarios with varying operating environments. Attached Figure Description

[0057] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0058] Figure 1 This is a schematic diagram illustrating an exemplary application scenario of an embodiment of this application;

[0059] Figure 2 This is a schematic diagram of the structure of a fault detection device for a smart water station according to an embodiment of this application;

[0060] Figure 3 This is a flowchart illustrating a fault detection method for a smart water station according to an embodiment of this application.

[0061] Figure 4 This is a schematic diagram illustrating the displacement data calculation principle of an embodiment of this application;

[0062] Figure 5 This is a schematic diagram of the buoy fault identification logic flow in an embodiment of this application;

[0063] Figure 6 This is a schematic diagram illustrating the calculation principle of the response interruption point in an embodiment of this application.

[0064] Reference numerals: 101, pontoon body; 102, guide rail; 103, vision acquisition component; 1031, first image sensor; 1032, second image sensor; 104, lifting module; 105, controller. Detailed Implementation

[0065] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0066] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0067] This application applies to smart water station systems with high structural integration, electrically adjustable water intake mechanisms, and operating environments with complex underwater resistance or risks of siltation. Application scenarios include, but are not limited to:

[0068] Small and medium-sized distributed water stations that use an electric lifting structure to control the depth of the pontoons;

[0069] Artificial wetland water stations or primary water source stations located in waters with dense aquatic plants and impurities;

[0070] The goal is to create an unmanned water station with video capture capabilities, enabling automatic fault identification and diagnosis.

[0071] It should be noted that the fault detection method proposed in this application mainly addresses two core technical problems in the automatic control process of existing water intake floats in water stations: motion feedback distortion and unclear fault location. It proposes a detection method that combines displacement data and image data, and introduces response interruption judgment and structural target recognition mechanisms.

[0072] Please see Figure 1 This figure is a schematic diagram of an exemplary application scenario provided by an embodiment of this application.

[0073] like Figure 1 As shown, water stations typically acquire current water level information through level sensors and input this information to a controller. The controller then generates a target displacement command based on the water level change, which controls the lifting structure to drive the float in the water to move up and down along the guide rail, thereby adjusting the water intake depth.

[0074] Figure 1 Furthermore, to provide feedback on the position of the pontoon, the system is equipped with an encoder connected to the lifting structure. The encoder is used to detect the number of rotations or displacement of the lifting motor and feeds the displacement data back to the controller as the basis for determining the actual displacement of the pontoon.

[0075] Those skilled in the art will understand that in the existing control methods, the controller equates the displacement fed back by the encoder to the actual movement of the pontoon body. However, in actual applications, when there is obstruction on the guide rail, underwater structure entanglement, or obstruction of the pontoon mechanism, the pontoon may fail to move according to the target command, while the encoder will still generate feedback of displacement due to the motor action. This leads to distortion in the control system's judgment of the pontoon's operating status, thereby affecting the water intake accuracy and safe operation of the entire water station.

[0076] Please see Figure 2 The figure is a schematic diagram of a fault detection device for a smart water station provided in an embodiment of this application. The fault detection device is used to detect the fault type of the float body 101, and the fault detection device includes:

[0077] Guide rail 102 is used to limit the lifting and lowering movement of the float body 101 along its length direction. The guide rail 102 may be a stainless steel frame, carbon fiber slide or other structural materials with water corrosion resistance and guiding function.

[0078] The lifting module 104 is used to drive the float body 101 in the water to be lifted and lowered along the guide rail 102 in response to the water level of the water station to be tested.

[0079] The lifting module 104 includes a water level sensor, a position detection component, and a lifting mechanism;

[0080] It is understood that the lifting mechanism may include electrically controlled lifting structures such as electric lead screws, gear chain mechanisms, and synchronous belt mechanisms. The position detection component integrates devices such as encoders or magnetic scales for detecting displacement during the lifting process. The water level sensor may be an ultrasonic water level sensor, a pressure water level sensor, a capacitive level gauge, or a radar water level sensor, etc., used to monitor water level information in real time and feed it back to the controller 105 so as to generate target control commands for the lifting and lowering of the float. The selection of the water level sensor can be flexibly adjusted according to the water quality conditions, installation space, and accuracy requirements of the water station's environment, and is not limited in this application.

[0081] The visual acquisition component 103 includes a first image sensor 1031 and a second image sensor 1032. The first image sensor 1031 is disposed in the guide rail 102 at a position perpendicular to the float body 101, and the second image sensor 1032 is disposed on the outside of the float body 101. The image sensors can be CMOS or CCD devices, and have industrial-grade resolution and frame rate requirements.

[0082] Controller 105 is connected to the lifting module 104 and the vision acquisition component 103, and controller 105 is configured as follows:

[0083] Obtain water level information and control the raising and lowering of the buoy;

[0084] Based on the displacement feedback of the lifting module 104 and the top image sequence of the first image sensor 1031, it is determined whether the conditions for entering the fault identification stage are met.

[0085] When the conditions for entering the fault identification stage are met, the response interruption point of the float body 101 is calculated, the target area of ​​the float body 101 is determined, and the second image sensor 1032 is controlled to acquire the target image.

[0086] Structural features are extracted from the target image and input into a preset fault identification network to output the fault type of the pontoon body 101.

[0087] The controller 105 can be a general-purpose embedded processing unit (such as STM32, ARM Cortex series control chips), an industrial PC, an FPGA platform, or an edge computing module with AI computing capabilities, and this application does not impose any restrictions.

[0088] Next, with reference to the accompanying drawings, a fault detection method for a smart water station provided in an embodiment of this application will be described. Figure 3The method shown is applied to a fault detection device, which includes a lifting module 104, a vision acquisition component 103, a guide rail 102, and a float body 101. The lifting module 104 is used to raise and lower the float body 101 in the water in response to the water level of the water station to be tested. The vision acquisition component 103 includes a first image sensor 1031 and a second image sensor 1032. The first image sensor 1031 is disposed in the guide rail 102 at a position perpendicular to the float body 101, and the second image sensor 1032 is disposed outside the float body 101. The method includes:

[0089] S1: Based on the acquired water level information, control the lifting module 104 to drive the float body 101 to move up and down along the guide rail 102;

[0090] In this embodiment, the lifting module 104 includes a water level sensor, a position detection component, and a lifting mechanism. The water level sensor is used to collect real-time liquid level data of the target water body, the position detection component is used to provide feedback on the actual displacement information of the float, and the lifting mechanism can realize the lifting control of the float based on an electric screw or gear chain structure.

[0091] S2: Determine whether the conditions for entering the fault identification stage are met based on the position detection component in the lifting module 104 and the first image sensor 1031;

[0092] In this embodiment, displacement feedback values ​​during the lifting process are collected along with a sequence of top images obtained by the first image sensor 1031. A preset spatial height mapping model is used to convert the changes in the top contour of the pontoon in the images into actual displacement, which is then compared and analyzed with the feedback from the displacement sensor. When the difference between the two types of displacement data exceeds a threshold, or when the image response trend is interrupted, it can be determined that the pontoon may be malfunctioning, thus meeting the conditions for entering the fault identification stage.

[0093] S3: If satisfied, the float fault identification logic is executed through the second image sensor 1032 to obtain the float fault type;

[0094] In this embodiment, when the fault identification stage is triggered, the controller 105 reverse-engineers the target area of ​​the pontoon structure based on the image response interruption point and controls the second image sensor 1032 to acquire images of the target area. The acquired images will be processed by algorithms such as image enhancement, edge detection, and contour segmentation to extract structural feature information, and then input into a fault identification network trained based on historical fault images of the pontoon, ultimately outputting a fault type label. Fault types may include, but are not limited to, pontoon jamming, guide rail 102 jamming, attitude tilting, or bottom floating obstruction.

[0095] It is readily understood that this application assumes that the lifting module 104 can stably drive the pontoon to move along the guide rail 102 after receiving the control command, and that the encoder or magnetic ruler-type position detection component can output the theoretical motion trajectory of the pontoon in real time; simultaneously, the first image sensor 1031 can stably acquire continuous images of the top contour of the pontoon under synchronous timing. Based on the above assumptions, the system can obtain consistent displacement feedback and image response under normal operating conditions.

[0096] In traditional buoy control systems for water stations, the buoy position is largely determined by displacement sensors (such as wire encoders or electric displacement feedback devices). However, when there is jamming in the guide rail 102, buoy deflection, or local resistance underwater (such as entanglement in weeds or corrosion of the guide rail), the system may still misjudge the situation due to the motor running without load, leading to a continuous feedback of an erroneous state indicating that the buoy is moving normally. In such scenarios, existing technologies often lack redundant channels to determine whether a movement is a false alarm, and even more so, they lack the logic to identify whether a local response interruption has occurred in a specific structural area of ​​the buoy. This results in fault diagnosis remaining at the level of anomaly occurrence rather than precise location of the anomaly source. Furthermore, traditional vision fusion systems often employ a full-structure image acquisition scheme, acquiring a complete image before performing full-image analysis. This results in high imaging redundancy, low recognition efficiency, and difficulty in ensuring image recognition stability in various complex underwater environments.

[0097] In this embodiment, a continuous image sequence of the top of the pontoon is acquired by introducing a first image sensor 1031, and a corresponding relationship is established with the displacement feedback information of the lifting module 104. The image response interruption point at the top of the pontoon is determined using optical flow method. This not only allows for real-time monitoring of the interruption sequence of the structural response but also provides precise anchor points in the time dimension for subsequent identification. Based on the response interruption point, and combined with the segment mapping model of the guide rail 102 structure, the target area can be further determined by reverse calculation on the pontoon structure, and the second image sensor 1032 can be controlled to image only this target area.

[0098] Next, the part of the method of this application regarding the lifting module 104 driving the float body 101 to rise and fall along the guide rail 102 will be further elaborated.

[0099] In traditional water intake systems, floats typically rely on the natural fluctuations of the water body to move up and down, using a follow-up hose or connecting rod to guide water samples. However, this passive floating mode is no longer sufficient for complex water station operations involving precise water intake, water layer selection, and water quality difference analysis. For example, in scenarios with multi-level water intake, frequent changes in reservoir water levels, or artificial adjustments to the intake depth, relying solely on the natural floating height of the float is insufficient to control the actual intake depth, easily leading to unclear intake layers, insufficient sample representativeness, or complete loss of intake capability at low water levels.

[0100] This application employs an electrically controlled lifting module 104 to rigidly drive and control the pontoon body 101. The pontoon's movement direction is limited by a guide rail 102, and displacement sensing is achieved using a height feedback device, thus enabling precise control of the pontoon's lifting height. Compared to traditional water level-dependent floating methods, this application's solution has the following advantages: First, it can actively respond to target water layer settings, allowing the pontoon to remain aligned with the target water intake position even under non-natural water level conditions; second, because the structure uses linear guide rails 102 or sliding rails to limit the pontoon's movement direction, undesirable movements such as attitude deflection, drift, and collisions can be effectively avoided, improving structural stability and repeatability.

[0101] In one example, the specific steps of S1 are as follows:

[0102] S1.1: Obtain the water level information collected by the water level sensor;

[0103] Specifically, the lifting module 104 is equipped with at least one set of water level sensors for real-time monitoring of the current water level of the water station under test.

[0104] Furthermore, in order to improve the accuracy of water level monitoring, the sensor output value can be filtered to remove interference caused by water surface fluctuations, thereby obtaining more stable water level reference information.

[0105] S1.2: Compare the water level information with the preset target water intake depth to determine the target position of the pontoon body 101;

[0106] Specifically, after obtaining the current water level, the controller 105 calculates the difference between the current water level and the target water intake layer depth according to the set target water intake layer logic. The target water intake layer depth can be set to a fixed depth or a depth calibration value that is dynamically adjusted according to water quality parameters; this application does not impose any limitation on this.

[0107] Furthermore, after calculating the target position, the controller 105 uses the target position as the desired position of the pontoon, generating a specific displacement command value. In particular, to avoid frequent minor adjustments, this application can set a position control dead zone threshold. The lifting mechanism is only activated for adjustment when the deviation between the current pontoon position and the target position exceeds the set position control dead zone threshold, thereby improving system stability.

[0108] How to generate specific displacement command values ​​based on the desired position of the pontoon is a matter of existing technology and will not be elaborated upon here.

[0109] S1.3: Control the lifting mechanism to move the float body 101 to the height corresponding to the target position;

[0110] Specifically, the controller 105 generates a PWM control signal or stepper motor drive pulse based on the displacement command value calculated in S1.2, and controls the motor movement. Simultaneously, to ensure the float is actually in position, position feedback can be provided by an encoder or magnetic scale, performing a closed-loop comparison between the actual lifting value and the target value. During the lifting process, if abnormal speed or sudden load changes are detected, the lifting process can be interrupted, triggering a fault warning or protection mechanism.

[0111] Next, we will further elaborate on the part of the method in this application regarding the determination of the pontoon position.

[0112] In existing technologies that use motors to drive objects, the theoretical feedback value of mechanical transmission is generally used as the main basis for determining whether the object's position has changed. However, in the application scenario of this application, especially when foreign objects are adsorbed at the bottom of the float, underwater guide rail 102 is obstructed, and water turbulence interferes, even if the float is stuck or completely stationary, the feedback data obtained by the position detection component still shows a normal movement trend because the drive motor is still moving. This misleads the system into believing that the float has moved according to the instructions, thus causing the system to fail to trigger the abnormal detection logic for a long time, ultimately resulting in a deviation in water intake accuracy or the water pump running dry.

[0113] This application introduces a first image sensor 1031 with a vertical viewing angle inside the guide rail 102 at the top of the pontoon, and designs an image displacement estimation auxiliary judgment path to determine whether the pontoon has actually undergone spatial displacement, thus serving as one of the core bases for entering the fault identification stage.

[0114] Understandably, the imaging direction is consistent with the lifting and lowering direction of the pontoon body 101, thus enabling the acquisition of image sequences of the pontoon's top area during the lifting and lowering process from a top-down perspective. The vertical arrangement ensures that the pontoon's vertical movement trajectory is accurately reflected in the scaling changes of the circular outline in the image. This allows for the stable conversion of pixel changes into the pontoon's actual vertical displacement value through a pre-defined spatial height mapping model. The use of a top-down perspective, rather than a traditional side view, is based on the pontoon's inherent high axisymmetry and the difficulty in imaging its side outline, thereby improving image response stability and measurement accuracy.

[0115] For example, displacement calculation can be referred to Figure 4 To understand, Figure 4 This is a schematic diagram illustrating the principle of displacement data calculation in an embodiment of this application, used to exemplarily explain the calculation process and differences between the first and second displacement data in buoy displacement detection.

[0116] Figure 4The diagram illustrates a position detection component used to encode and provide feedback on the displacement process of the lifting module 104, outputting corresponding first displacement data. This data represents the theoretical travel calculated internally by the control system based on displacement commands and encoder feedback, exhibiting high responsiveness and real-time performance. However, since it is based solely on drive-end feedback, it is susceptible to factors such as structural jamming and slippage, potentially deviating from the actual pontoon state.

[0117] Figure 4 The diagram further illustrates how the first image sensor 1031 captures a sequence of images of the top of the pontoon at a fixed acquisition period, and extracts the top contour feature sequence using an edge detection algorithm. In the figure, three different frame images demonstrate the contour scaling effect of the pontoon at three different times. The number of specific frame images can be set independently; the contour shows a clear shrinking trend as the pontoon descends. Using a spatial height mapping model, the pixel size of the pontoon's top contour in the image frames is mapped to the actual vertical height change, thus obtaining the second displacement data.

[0118] It should be noted that, since the image feedback path more directly reflects the actual state of the float body 101, the second displacement data is more accurate than the first displacement data in practical applications, especially in cases of structural anomalies. In this application, the second displacement data is used by default as the evaluation standard for the actual motion state, while the first displacement data is used as a control link reference. The two constitute a comparison benchmark to determine whether a fault response interruption has occurred.

[0119] It is understandable that, since encoder feedback is usually continuous and high-frequency, while image acquisition is limited by frame rate and processing cycle, there are natural differences between the two in terms of data length, sampling period and response granularity. This is one of the reasons why the two displacement data are stored in different data buffer structures as shown in the figure.

[0120] In one example, the specific steps of S2 are as follows:

[0121] S2.1: Obtain the first displacement data output by the position detection component;

[0122] Specifically, the position detection component is composed of an encoder sensor, which is installed on the drive shaft or related driven structure of the lifting mechanism. It collects displacement information of the float as it moves along the guide rail 102 by reading changes in rotational angle or linear displacement. The first displacement data reflects the theoretical displacement path of the float under ideal unobstructed movement conditions, and is a direct quantitative feedback between the control signal and physical execution.

[0123] In this embodiment, the encoder can be an incremental or absolute rotary encoder, which achieves discrete acquisition of displacement through high-resolution signal pulse feedback. Alternatively, a linear displacement sensor with a magnetic scale can be used to adapt to different types of guide rail 102 structures. The encoder output signal is converted into numerical form by the acquisition module and sent to the controller 105 for cross-comparison with subsequent image data.

[0124] S2.2: Acquire the top image sequence collected by the first image sensor 1031, wherein the acquisition period of the first image sensor 1031 is set synchronously with the operation period of the lifting module 104;

[0125] Specifically, the first image sensor 1031 is rigidly mounted on the vertical observation position of the guide rail 102 structure, with its optical axis pointing vertically to the top of the pontoon, so as to capture a planar view of the top of the pontoon.

[0126] Furthermore, in order to ensure that the image sequence can completely cover the changing trajectory of the float throughout the entire lifting cycle, the acquisition cycle of the first image sensor 1031 is set to be synchronized with the action cycle of the lifting module 104. That is, whenever the lifting module 104 completes a unit control cycle, the image sensor triggers a shooting action until the displacement detection component reports the completion of the lifting action and captures the last top image, thereby forming a top image sequence.

[0127] S2.3: Perform edge detection on the top images in the top image sequence sequentially to obtain a contour feature sequence;

[0128] Specifically, to extract quantifiable motion features from consecutive image frames, this embodiment employs an edge detection algorithm to process the top image sequence. The purpose of edge detection is to accurately identify the boundary position of the pontoon's top contour and extract its two-dimensional projection size in the image frame, providing a reliable basis for subsequent pixel-to-real-space mapping. In practical applications, preferred edge detection algorithms include the Canny edge operator or the Sobel operator to adapt to different image contrast and noise environments under varying shooting conditions.

[0129] Those skilled in the art will understand that how to extract contour features through edge detection is an existing technology, such as extracting the pixel size values ​​of its circumscribed rectangle or circumscribed circle, which will not be elaborated here.

[0130] S2.4: Based on a pre-defined spatial height mapping model, the pixel size changes of the contour feature sequence in the image are converted into vertical displacement values ​​to obtain second displacement data, wherein the spatial height mapping model is constructed based on image projection transformation relationships;

[0131] Specifically, to convert the contour feature sequence in the image into a meaningful displacement, this embodiment introduces a spatial height mapping model. The spatial height mapping model is established based on the projection geometry under fixed viewing conditions, fitting and mapping the pixel size of the contour in the image to the physical distance between the top of the pontoon and the image sensor. In this embodiment, physical calibration is used to sample images of the pontoon at different height positions, and a calibration curve or interpolation model between pixel size and height is established.

[0132] In this embodiment, the contour dimensions of each frame of the image are calculated in reverse using a spatial height mapping model to obtain the corresponding vertical displacement value, and finally a second displacement data synchronized with the image frame is constructed.

[0133] Furthermore, the spatial height mapping model can also have scalability and recalibration capabilities, based on the first image sensor 1031 with different specifications or different installation heights.

[0134] In one example, the criteria for determining whether to enter the fault identification stage based on displacement data include the following two aspects:

[0135] Firstly, the judgment is based on the absolute value of the difference between the first displacement data and the second displacement data.

[0136] In one scenario, if the absolute value of the difference between the first displacement data and the second displacement data is greater than or equal to a preset error threshold, it can be determined that the fault identification stage has been entered. The error threshold can be obtained by retrospectively statistically analyzing the buoy lifting and lowering process in historical operating data.

[0137] Preferably, the maximum deviation between the encoder feedback displacement and the image measurement displacement is extracted under multiple fault-free conditions and used as an initial empirical threshold, which can then be dynamically fine-tuned based on on-site feedback.

[0138] Understandably, when the float body 101 experiences unexpected interference such as structural jamming, attitude deviation, or obstruction by surface debris during its raising and lowering process, its motion behavior in the image space often cannot perfectly match the encoder feedback response curve, resulting in a significant difference in displacement estimation between the two. In this case, the true response reflected in the image will lag or even be lost, while the encoder signal continues to output the ideal displacement, thus constituting an abnormal deviation. By setting an error threshold to trigger the judgment of the deviation, potential mechanical faults or structural malfunctions in the system can be detected in a timely manner.

[0139] In another scenario, if the absolute value of the difference between the first displacement data and the second displacement data is less than a preset error threshold, there is no need to proceed to the fault identification stage. In this case, it can be assumed that the current movement trend of the pontoon in the lifting path remains stable, the visual response and the electronic control execution path have a good synchronization relationship, and there are no abnormal signs such as drifting, jamming, or misalignment. Therefore, there is no need to perform deeper structural image analysis.

[0140] Understandably, when the float moves smoothly within the lifting guide rail 102, the mechanical structure is in normal condition, the sensors are working stably, and the lighting and image quality meet expectations, the system will exhibit a high degree of overlap between the encoder feedback path and the image estimation path. At this point, even minor errors can be tolerated through a reasonable error threshold setting, thus avoiding frequent entry into the image recognition stage due to slight errors, saving computational resources, and improving overall recognition efficiency and stability.

[0141] Secondly, the judgment is made based on the changes in the second displacement data within the action cycle.

[0142] In one scenario, when the second displacement data does not change or the rate of change is lower than the set change threshold within the action cycle, it can be determined that the fault identification stage has been entered. The change threshold can be obtained by statistically fitting a large number of pontoon lifting processes under normal working conditions, and then optimized based on the motion inertia of the pontoon body 101 and the image sampling frequency.

[0143] It is understandable that when the float body 101 is executing the lifting command, if the top contour features reflected in the top image sequence remain still or only experience very slight shaking, it means that the float has not actually completed the action as expected. Since the first image sensor 1031 is always located on top of the guide rail 102, its continuous monitoring of the top image contour can sensitively detect motion stagnation, thereby using insufficient rate of change as an effective trigger mechanism for entering the next stage of judgment.

[0144] In another scenario, when the rate of change of the second displacement data within the action cycle is greater than or equal to a set change threshold, there is no need to proceed to the fault identification stage. In this case, it can be assumed that the pontoon's response behavior in the image space is consistent with the lifting command, and its motion trend does not show obvious signs of structural abnormalities or action failure.

[0145] Understandably, when the top contour in the top image sequence exhibits continuous and stable scale changes, it indicates that the pontoon does indeed move with the lifting mechanism, the system structure is not disturbed, and the lifting control closed loop can be considered normal. In this case, even if there are minor errors in the encoder or other components, interference signals can still be eliminated through the dynamic response of the displacement change rate, avoiding misjudgment.

[0146] Furthermore, the first and second aspects can be combined to determine whether to enter the fault identification stage.

[0147] In determining whether to enter the fault identification stage by combining the first and second aspects, if the results based on the first and second aspects are inconsistent, the results based on the second aspect take precedence. That is, the second aspect has a higher priority than the first aspect. This is because the second aspect, based on the image change trend, directly reflects the actual response behavior of the pontoon top. Its stability and anti-interference capabilities are relatively strong. Especially when the encoder experiences zero-point drift, sampling lag, local failure, or errors caused by external vibration, the first displacement data may deviate, but the vertical trend of the image contour can still accurately reflect the dynamic state of the pontoon. The image response is essentially a projection of the pontoon's physical movement into the visual domain. Its change trend can form a continuous trajectory and is constrained by the fixed viewing angle of the first image sensor 1031, making it less susceptible to external occasional disturbances.

[0148] In addition to the two aspects mentioned above, this application embodiment may also determine whether to enter the fault identification stage through other methods or in combination with other methods.

[0149] For example, a judgment can be made based on the direction of change of the second displacement data and the direction of change of the first displacement data.

[0150] Optionally, when the direction of change of the second displacement data is opposite to the direction of change of the first displacement data, it can be determined that the actual response of the float in the current action cycle is abnormal, meeting the conditions for entering the fault identification stage. For example, when the lifting command is upward, the first displacement data recorded by the encoder shows an upward trend, but the second displacement data calculated from the image sequence shows a downward trend, indicating that the actual movement direction of the float deviates from the theoretical command. This may be caused by the guide rail 102 jamming, pulley slippage, abnormal force on the float, etc., and further fault type needs to be located.

[0151] Conversely, when the two directions are consistent and fluctuate within a reasonable threshold range, even with a small error, the buoy response can be considered normal, and there is no need to enter the fault identification stage. Directional consistency, as a supplementary means to trend matching, can effectively avoid boundary situations that cannot be judged due to small displacement, such as the micro-motion state when the buoy has just started or is nearing its termination position.

[0152] Next, we will further elaborate on the part of the method for identifying pontoon faults in this application.

[0153] Those skilled in the art will understand that during the long-term operation of the pontoon lifting structure, due to the interaction of factors such as the water environment, wear of the guide rail 102, entanglement with aquatic plants, and structural deformation, common faults include: stroke interruption caused by partial jamming of the guide rail 102, attitude deviation caused by deformation of the pontoon shell, suspension obstruction caused by bottom confinement, and non-uniform motion caused by loose connecting parts. These faults not only occur in different areas of the pontoon structure, but also correspond to different stages of the lifting process.

[0154] Taking a typical scenario as an example, if the bottom of the buoy is restricted due to weeds or sediment accumulation, the abnormality usually manifests as an abnormal structural response only at the end of the descent; while if there is metal fatigue or foreign object jamming in the middle section of the buoy guide rail 102, obvious reverse resistance can be triggered as soon as the movement starts, which manifests as a serious discrepancy between the upper displacement feedback and the image response.

[0155] Therefore, this application believes that the time point of occurrence of pontoon structure abnormalities and their corresponding height often have strong spatial orientation. In other words, the time of occurrence of different faults implicitly corresponds to the spatial distribution of the abnormal parts of the structure.

[0156] For reference Figure 5 To understand, Figure 5 This is a schematic diagram of the buoy fault identification logic flow in an embodiment of this application.

[0157] In one example, the specific steps of S3 are as follows:

[0158] S3.1: Calculate the response interruption point based on the feedback information of the lifting module 104 during the current operation cycle and the detection information of the first image sensor 1031;

[0159] It is understandable that the response interruption point is not an absolute coordinate on the structural position, but rather refers to the starting segment within a certain lifting and lowering motion cycle where the image change trend and the theoretical displacement trend first become disconnected. Because image information has higher realism, the response interruption point marks the location where the structure first loses its effective linkage during the current motion, thus serving as a key focusing area for subsequent visual inspection.

[0160] Furthermore, the response interruption point not only has a clear temporal attribute (corresponding to the frame number in the image sequence), but also naturally carries high spatial semantics. The theoretical displacement segment corresponding to it can be mapped to a specific area of ​​the pontoon structure. This characteristic allows the second image sensor 1032 to no longer perform a full-field indiscriminate scan when entering the fault identification stage, but instead to acquire target images directionally based on the structural area located by the response interruption point.

[0161] Taking the response to the interruption point as an example, please refer to Figure 6 To understand, Figure 6This is a schematic diagram illustrating the calculation principle of the response interruption point in an embodiment of this application.

[0162] Figure 6 The diagram illustrates how, firstly, the theoretical motion trajectory of the pontoon is constructed based on the target displacement command and feedback displacement data of the lifting module 104 within the current operating cycle. This trajectory reflects the continuous displacement of the pontoon within the guide rail 102 under ideal operating conditions. Subsequently, by combining the calibration parameters and viewing angle configuration of the first image sensor 1031, the theoretical motion trajectory is mapped to the image space through image back projection, yielding the theoretical change area of ​​the pontoon's top in the image sequence.

[0163] Understandable, Figure 6 Although only a theoretical change area at the top of the pontoon is marked, in actual operation, this theoretical change area is a series of regions generated frame-by-frame on the entire displacement time axis based on the continuous target displacement commands of the pontoon. Each theoretical change area corresponds to the top contour shape of the pontoon in image space at a certain theoretical height, reflecting how the image contour of the pontoon should change over time under different displacement states. Therefore, Figure 6 The single theoretical change region shown is merely a schematic representation of a specific frame in the image sequence, used to intuitively illustrate the principle. Overall, the image backprojection process is based on dynamic modeling results of temporal continuity, and its output is a multi-frame theoretical image contour evolution sequence.

[0164] Figure 6 Furthermore, it is shown that the actual change area of ​​the pontoon top is extracted from the actual acquired top image sequence using image processing techniques such as edge detection and contour analysis. The actual change area reflects the actual image response behavior of the pontoon top during the lifting and lowering process. Due to abnormal factors such as structural jamming and interference from guide rail 102, there may be missing or interrupted responses.

[0165] Furthermore, optical flow is used to compare the response trends between the theoretical and actual change regions frame by frame to detect the cumulative trend of response offset. At the beginning of a certain image frame, if the actual response image no longer changes as expected with the theoretical displacement, i.e., the response trend is interrupted or significantly lagging, the control command segment corresponding to that frame can be identified, which is the response interruption point.

[0166] In one example, the steps for calculating the response interruption point are as follows:

[0167] S3.1.1: Obtain the target displacement command and displacement feedback information of the lifting module 104 in the current action cycle, and construct the theoretical motion trajectory of the float body 101 in the current action cycle;

[0168] It is understood that the target displacement command originates from the controller 105, and the displacement feedback information originates from the displacement detection component. How the theoretical motion trajectory is constructed is a matter of existing technology. For example, based on the temporal alignment of the displacement command and the feedback displacement, a target-feedback motion lookup table based on timestamps can be constructed. Furthermore, by compensating for the response time characteristics of the lifting module 104 (such as acceleration delay, braking time, etc.), a detailed model of the expected movement behavior of the pontoon within the current cycle can be achieved, ultimately generating a temporally continuous and segmentally traceable theoretical motion trajectory. This will not be elaborated upon here.

[0169] S3.1.2: Based on the viewing angle parameters and calibration parameters of the first image sensor 1031, the theoretical motion trajectory is back-projected in the image space to obtain the theoretical change area of ​​the top of the pontoon;

[0170] In this embodiment, the viewing parameters include, but are not limited to, focal length, principal point position and viewing angle, and the calibration parameters include, but are not limited to, distortion correction matrix, intrinsic parameter matrix and pose information.

[0171] Those skilled in the art will understand that, given the viewpoint and calibration parameters, a projection transformation matrix from three-dimensional space to a two-dimensional image plane can be constructed. Based on this, each discrete position point in the theoretical motion trajectory is used as input, and the edge contour range in the image frame corresponding to each theoretical position of the pontoon top is calculated through back-projection transformation.

[0172] Furthermore, to adapt to the actual situation where the top structure of the pontoon has a certain curvature or specular reflection effect, a set of compensation parameters can be introduced into the projection operation, such as the geometric boundary expansion factor obtained based on the top structure modeling, to dynamically adjust the size range of the theoretical change area, improve the fault tolerance of the theoretical area matching the actual image, and finally output the theoretical change area in an image sequence.

[0173] S3.1.3: The theoretical change area is compared with the actual change area of ​​the top of the pontoon in the detection information by optical flow method to detect the change in response trend between consecutive image frames, so as to obtain the start time of the interruption of the image response at the top of the pontoon, and the control segment corresponding to the start time is taken as the response interruption point.

[0174] S3.2: Based on the response interruption point, determine the target area of ​​the float body 101, and obtain the target image of the target area through the second image sensor 1032;

[0175] Based on the response interruption point, the target area of ​​the float body 101 is determined, including:

[0176] The movement path of the pontoon body 101 along the guide rail 102 is divided into multiple preset structural sections, wherein each preset structural section corresponds one-to-one with the structural area of ​​the pontoon body 101.

[0177] It is understandable that the overall structure of the pontoon body 101 is composed of multiple functional modules or structural units connected longitudinally, including but not limited to: a top guide cap, a main buoy cavity section, a counterweight connecting section, and a bottom guide rail limiting structure. Based on the spatial position and distribution characteristics of these physical structures, the guide rail 102 can be divided into several preset structural sections along its length. The height range of each section is completely consistent with the size of its corresponding structural unit or has a precision tolerance setting, and the section division boundary can be accurately determined by calibrating the pontoon structure design drawings. In actual deployment, the number and position of the preset structural sections do not need to be generated in real time, but are stored in the controller 105 in the form of a configuration file during the system initialization phase for subsequent section identification and acquisition logic mapping.

[0178] Furthermore, once the response interruption point is calculated, its corresponding control segment can be located. This control segment refers to the displacement range of the pontoon corresponding to the response interruption point in the theoretical motion trajectory, and has a unique identifier, such as a segment number or height label. The controller 105 can quickly locate the target structural area corresponding to the response interruption point by matching the numbering or identifier mapping relationship between the control segment and the preset structural segment. This target structural area is the target of image acquisition, representing the location where the pontoon body 101 is most likely to experience structural abnormalities.

[0179] The corresponding preset structural segment is determined based on the control segment of the response interruption point;

[0180] The structural area of ​​the corresponding pontoon body 101 of the corresponding preset structural section is taken as the target area.

[0181] It should be noted that, to ensure effective acquisition of the target image, the second image sensor 1032 is configured on the outside of the pontoon, possessing an imaging angle along the pontoon's axis, and supporting automatic acquisition and buffering of image frames as the target area passes through its imaging field of view. In an optional embodiment, the second image sensor 1032 can work with a limit trigger or a time synchronizer to automatically activate the shutter action when the target area passes through, ensuring consistency between image acquisition accuracy and response point recognition time.

[0182] S3.3: Perform image enhancement, edge detection, and contour segmentation on the target image to obtain structural features;

[0183] Those skilled in the art will understand that how to identify and segment the target area after obtaining the image is prior art, and this application will not elaborate on it here.

[0184] It should be noted that the structural features in this application refer to the set of image information extracted after multi-layer processing of the target area image of the pontoon body 101, which can characterize the geometric structure, contour morphology, and local deformation state of the area. These structural features not only reflect the visual information of the pontoon surface or boundary, but also include distortion patterns, crack trends, jamming marks, or other microscopic defect contours that may appear on the pontoon structure under abnormal stress or deformation conditions, supporting subsequent fault type identification operations.

[0185] S3.4: Input the structural features into a preset fault identification network, process the structural features through the fault identification network, and output the buoy fault type, wherein the fault identification network is trained using historical fault images of the buoy;

[0186] In one example, the fault identification network includes a structure encoding layer, a state reasoning layer, and a fault classification output layer, wherein:

[0187] The structural coding layer is used to vectorize the input structural features, including edge continuity parameters, contour closure index, and occlusion area ratio.

[0188] The state reasoning layer is equipped with a graph structure attention mechanism, which is used to establish spatial correlations between structure encoding vectors and reason about the integrity state and stress consistency characteristics of the pontoon structure.

[0189] The fault classification output layer is used to output the fault type of the pontoon through the Softmax classifier based on the integrity status and force consistency characteristics. The fault types include pontoon jamming, guide rail 102 jamming, bottom floating obstruction, and pontoon attitude tilting.

[0190] In one example, this application provides a fault detection system for a smart water station, applied to a fault detection device. The fault detection device includes a lifting module 104, a vision acquisition component 103, a guide rail 102, and a float body 101. The lifting module 104 is used to lift and lower the float body 101 in the water in response to the water level of the water station to be detected. The vision acquisition component 103 includes a first image sensor 1031 and a second image sensor 1032. The first image sensor 1031 is disposed in the guide rail 102 at a position perpendicular to the float body 101, and the second image sensor 1032 is disposed outside the float body 101. The system includes:

[0191] The water level control module is used to acquire the water level information of the water station to be tested, and control the lifting module 104 to move the float body 101 to the height corresponding to the target position according to the water level information.

[0192] The displacement determination module is used to determine whether the conditions for entering the fault identification stage are met based on the first displacement data output by the position detection component in the lifting module 104 and the top image sequence collected by the first image sensor 1031.

[0193] The image acquisition module is used to control the second image sensor 1032 to acquire the target image of the float body 101 after the conditions for entering the fault identification stage are met.

[0194] The feature extraction module is used to perform image enhancement, edge detection, and contour segmentation on the target image to obtain the structural features of the pontoon;

[0195] The fault identification module is used to input the structural features into a preset fault identification network and output the corresponding float fault type based on the fault identification network.

[0196] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A fault detection method for a smart water station, applied to a fault detection device, the fault detection device comprising a lifting module, a vision acquisition component, a guide rail, and a float body, wherein the lifting module is used to lift and lower the float body in the water in response to the water level of the water station to be detected, the vision acquisition component comprising a first image sensor and a second image sensor, the first image sensor being disposed in the guide rail perpendicular to the float body, and the second image sensor being disposed on the outside of the float body, characterized in that... The method includes: Based on the acquired water level information, the lifting module is controlled to drive the float body to rise and fall along the guide rail; The system determines whether the conditions for entering the fault identification stage are met based on the position detection component in the lifting module and the first image sensor. If satisfied, the float fault identification logic is executed through the second image sensor to obtain the float fault type. The float fault identification logic locates the target area of ​​the float body based on the response interruption point and performs structural feature extraction and fault type identification processing on the target area. The pontoon fault identification logic includes: Calculate the response interruption point based on the feedback information of the lifting module during the current action cycle and the detection information of the first image sensor; Based on the response interruption point, the target area of ​​the pontoon body is determined, and a target image of the target area is obtained through the second image sensor; The target image is subjected to image enhancement, edge detection, and contour segmentation to obtain structural features; The structural features are input into a preset fault identification network, which processes the structural features and outputs the buoy fault type. The fault identification network is trained using historical fault images of the buoy. The step of calculating the response interruption point based on the feedback information within the current operation cycle of the lifting module and the detection information from the first image sensor includes: Obtain the target displacement command and displacement feedback information of the lifting module in the current action cycle, and construct the theoretical motion trajectory of the float body in the current action cycle; Based on the viewing angle parameters and calibration parameters of the first image sensor, the theoretical motion trajectory is back-projected in the image space to obtain the theoretical change area at the top of the pontoon. The theoretical change region is compared with the actual change region of the top of the pontoon in the detection information using optical flow method to detect the change in response trend between consecutive image frames, so as to obtain the start time of the interruption of the image response at the top of the pontoon, and the control segment corresponding to the start time is taken as the response interruption point.

2. The fault detection method for a smart water station according to claim 1, characterized in that, Based on the response interruption point, the target area of ​​the pontoon body is determined, including: The movement path of the pontoon body along the guide rail is divided into multiple preset structural sections, wherein each preset structural section corresponds one-to-one with the structural area of ​​the pontoon body. The corresponding preset structural segment is determined based on the control segment of the response interruption point; The structural area of ​​the corresponding pontoon body in the corresponding preset structural section is taken as the target area.

3. The fault detection method for a smart water station according to claim 1, characterized in that, The fault identification network includes a structure encoding layer, a state reasoning layer, and a fault classification output layer, wherein: The structural coding layer is used to vectorize the input structural features, including edge continuity parameters, contour closure index, and occlusion area ratio. The state reasoning layer is equipped with a graph structure attention mechanism, which is used to establish spatial correlations between structure encoding vectors and reason about the integrity state and stress consistency characteristics of the pontoon structure. The fault classification output layer is used to output the fault type of the buoy through the Softmax classifier based on the integrity status and force consistency characteristics. The fault types include buoy jamming, guide rail jamming, bottom floating obstruction, and buoy attitude tilting.

4. The fault detection method for a smart water station according to claim 3, characterized in that, The lifting module includes a water level sensor, a position detection component, and a lifting mechanism. Controlling the lifting module to drive the float body to rise and fall along the guide rail based on the acquired water level information includes: Obtain the water level information collected by the water level sensor; The water level information is compared with the preset target water intake depth to determine the target position of the pontoon body; The lifting mechanism is controlled to move the pontoon body to the height corresponding to the target position.

5. The fault detection method for a smart water station according to claim 1, characterized in that, The system determines whether the conditions for entering the fault identification stage are met based on the position detection component in the lifting module and the first image sensor, including: Obtain the first displacement data output by the position detection component; The top image sequence acquired by the first image sensor is obtained, wherein the acquisition period of the first image sensor is set synchronously with the action period of the lifting module; Edge detection is performed sequentially on the top images in the top image sequence to obtain a contour feature sequence; According to a pre-defined spatial height mapping model, the pixel size changes of the contour feature sequence in the image are converted into vertical displacement values ​​to obtain second displacement data, wherein the spatial height mapping model is constructed based on the image projection transformation relationship.

6. The fault detection method for a smart water station according to claim 5, characterized in that, The conditions for entering the fault identification stage include at least one of the following: The absolute value of the difference between the first displacement data and the second displacement data is greater than or equal to a preset error threshold. The second displacement data did not change during the operation cycle; The direction of change of the second displacement data is opposite to the direction of change of the first displacement data.

7. A fault detection system for a smart water station, used to implement the fault detection method for a smart water station as described in any one of claims 1 to 6, the system being applied to a fault detection device, the fault detection device comprising a lifting module, a vision acquisition component, a guide rail, and a float body, the lifting module being used to lift and lower the float body in the water in response to the water level of the water station to be detected, the vision acquisition component comprising a first image sensor and a second image sensor, the first image sensor being disposed in the guide rail at a position perpendicular to the float body, and the second image sensor being disposed on the outside of the float body, characterized in that... The system includes: The water level control module is used to acquire the water level information of the water station to be tested, and control the lifting module to move the float body to the height corresponding to the target position according to the water level information; The displacement determination module is used to determine whether the conditions for entering the fault identification stage are met based on the first displacement data output by the position detection component in the lifting module and the top image sequence collected by the first image sensor. The image acquisition module is used to control the second image sensor to acquire target images of the target area of ​​the float body after the conditions for entering the fault identification stage are met. The feature extraction module is used to perform image enhancement, edge detection, and contour segmentation on the target image to obtain the structural features of the pontoon; The fault identification module is used to input the structural features into a preset fault identification network and output the corresponding float fault type based on the fault identification network.

8. A fault detection device for a smart water station, used to implement the fault detection method for a smart water station as described in any one of claims 1 to 6, characterized in that, The fault detection device is used to detect the type of fault in the pontoon body, wherein the fault detection device includes: Guide rails are used to limit the vertical movement of the pontoon body along its length; The lifting module is used to drive the float body in the water to rise and fall along the guide rail in response to the water level of the water station to be tested. The visual acquisition component includes a first image sensor and a second image sensor. The first image sensor is disposed in the guide rail at a position perpendicular to the pontoon body, and the second image sensor is disposed on the outside of the pontoon body. A controller, connected to the lifting module and the vision acquisition component, is configured as follows: Obtain water level information and control the raising and lowering of the buoy; Based on the displacement feedback of the lifting module and the top image sequence of the first image sensor, it is determined whether the conditions for entering the fault identification stage are met. When the conditions for entering the fault identification stage are met, the buoy response interruption point is calculated, the target area of ​​the buoy body is determined, and the second image sensor is controlled to acquire the target image. Structural features are extracted from the target image and input into a preset fault identification network to output the fault type of the pontoon.

Images