Glass curtain wall loosening disease intelligent detection device and method based on cleaning operation

Abstract

The present application relates to a kind of glass curtain wall loosening disease intelligent detection device based on cleaning operation, including glass curtain wall cleaning tool and being set on glass curtain wall cleaning tool contact detection module, pose detection module, data processing storage transmission module and data processing module;Contact detection module is used to measure the distance between cleaning tool and glass curtain wall, threshold value is designed in combination with cleaning tool structure parameter, detect whether to enter cleaning operation link;Pose detection module is used to obtain the pose information of equipment, so as to obtain the position information of cleaning contact;Data processing storage transmission module is used to realize the transmission of pose detection data to data processing module;Data processing module is used to analyze and process the data collected by contact detection module and pose detection module, realize the distributed detection of glass curtain wall loosening condition.The present application can detect the loosening condition of existing glass curtain wall in real time, realize the timely judgment and acquisition of glass curtain wall disease problem.

Description

Technical Field

[0001] This invention relates to existing glass curtain wall inspection technologies, and more specifically, to an intelligent detection device and method for loosening defects in glass curtain walls based on cleaning operations. Background Technology

[0002] A curtain wall is the exterior cladding of a building, a lightweight wall structure with decorative effects commonly used in modern large and high-rise buildings. It consists of a structural frame and inlaid panels, and does not bear the load or action of the main structural element. In the main components of a glass curtain wall, glass is the most important material, and its performance largely determines the overall performance of the curtain wall. Currently, glass curtain walls dominate the building curtain wall industry; according to data from Curtain Wall Network, glass curtain walls account for 41% of all building curtain wall projects. Glass is currently the most widely used building material, accounting for over 70% of market demand.

[0003] While existing curtain wall inspection technologies have proposed various theoretical methods, most rely on sampling for safety testing. This sampling approach is difficult to apply to the routine maintenance of existing glass curtain walls and hinders a comprehensive assessment of their safety status. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent detection device and method for loosening defects in glass curtain walls based on cleaning operations, which can realize timely judgment and acquisition of defects in glass curtain walls while cleaning operations are being carried out on existing glass curtain walls of buildings.

[0005] The technical solution adopted by the present invention to solve its technical problem is: to construct an intelligent detection device for loosening defects of glass curtain walls based on cleaning operations, including glass curtain wall cleaning tools and contact detection module, pose detection module, data processing, storage and transmission module and data processing module set on the glass curtain wall cleaning tools;

[0006] The contact detection module is used to calculate the distance between the cleaning tool and the glass curtain wall, and to design a threshold based on the structural parameters of the cleaning tool to detect whether the cleaning operation has started.

[0007] The pose detection module is used to obtain the pose information of the device, thereby obtaining the position information of the cleaning contact point;

[0008] The data processing, storage, and transmission module is used to transmit pose detection data to the data processing module.

[0009] The data processing module is used to analyze and process the data collected by the contact detection module and the pose detection module to realize distributed detection of glass curtain wall loosening.

[0010] According to the above scheme, the contact detection module includes an ultrasonic distance sensor, which acquires the distance between the sensor on the device and the curtain wall glass based on ultrasonic ranging.

[0011] According to the above scheme, the pose detection module is an IMU attitude sensor, which includes an accelerometer, an angular velocity meter, and a magnetometer.

[0012] According to the above scheme, the data processing, storage and transmission module is connected to the pose detection module through a wire. The data processing, storage and transmission module is used to realize the wired transmission of pose detection data to the data processing module. After completing the data processing and storage, the data is sent through wireless communication via Bluetooth or the Internet.

[0013] According to the above scheme, the glass curtain wall cleaning tool is a cleaning scraper.

[0014] According to the above scheme, the ultrasonic distance sensor is located at the lower outer part of the cleaning squeegee tip, the pose detection module is located at the upper outer part of the cleaning squeegee tip, and the data processing, storage and transmission module is located at the upper outer part of the cleaning squeegee tip and behind the pose detection module.

[0015] This invention also provides an intelligent detection method for loosening defects in glass curtain walls based on cleaning operations. The method employs an intelligent detection device for loosening defects in glass curtain walls based on cleaning operations, and includes the following steps:

[0016] S1. Before carrying out cleaning operations, initialize the global coordinate system: take the three directions of northeast, south, and sky as the x, y, and z axes of the coordinate system, and take the position of the point at the start of the detection as the origin of the coordinate system.

[0017] S2, with period τ i The system collects detection data from the contact detection module and the pose detection module, and then transmits the collected data to the data processing module through the data processing, storage and transmission module.

[0018] S3, the data processing module integrates the displacement vectors of each cycle. The sensor's detection center position in the initial coordinate system is obtained during the i-th cycle. for

[0019] S4. The data processing module combines the location information of the collection points at all times and uses the least squares method to obtain the fitted glass surface plane.

[0020] S5. The data processing module collects multiple sets of characteristic parameters corresponding to existing and loose glass and characteristic parameters corresponding to loose glass, establishes a three-layer fully connected neural network for classification, and after the neural network is established and supervised model training is completed, the neural network is deployed on the device to realize distributed detection of glass curtain wall loosening.

[0021] According to the above scheme, in step S1, the contact detection sensor detects the distance D between the contact detection device and the curtain wall glass. When the distance D is less than the set threshold [D], that is, D < [D], the pose detection module sensor starts to collect data.

[0022] According to the above scheme, in step S2, the instantaneous acceleration at the sensor coordinates at the corresponding period time is collected by the accelerometer, angular velocity meter, and magnetometer. instantaneous angular velocity and angles in the geomagnetic field coordinate system The angles are initialized by converting the obtained angles relative to the magnetic field into angles lower than those in the initial coordinate system, i.e.:

[0023] For any different time t i The sensor direction is below, and the predicted value is obtained by accumulating the instantaneous angular velocity gyro, that is: the t-th... i Instantaneous angular velocity in the coordinate system at time t Transform to the t-th node through rotation. i-1 angular velocity in the time coordinate system have in, For the tth i Time t i-1 The rotation coordinate transformation matrix at time t; for the t-th time... i-1 Time t i-2 At any time, there is Then for any time t i Its instantaneous acceleration in the initial coordinate system is Then at any time t i The angle prediction value is obtained by accumulating the instantaneous angular velocity gyro:

[0024] Considering the different error models between predicted and observed values, if the accumulated angular velocity is used as the predicted value and the angle obtained from the magnetic field is used as the observed value, then it is assumed that the sensor at any time t... i The angle below is Where, k re and k obs For the weights of the predicted and observed values ​​relative to the true values, we have k. pre +k obs=1; Simultaneously, the actual value is used as a correction to the predicted value at that moment, resulting in...

[0025] For the change in displacement, at any time t i Below, there is Then there is Obtain the displacement vector in the initial coordinate system at each moment.

[0026] According to the above scheme, in step S3, considering the sensor position on the intelligent detection cleaning tool and the parameter settings at the contact point with the glass, that is, in the sensor coordinate system, the coordinates of the two ends of the contact point are [x0 y0 z0] and [-x0 y0 z0]. The line segment at the contact point is sampled into five points according to its length, namely [x0 y0 z0], [x0 / 2 y0 z0], [0 y0 z0], [-x0 / 2 y0 z0], and [-x0 y0 z0]. Combined with the direction vector obtained in step S2, the corresponding time t on the plane in the initial coordinate system is obtained. i The coordinates of each collection point are shown below.

[0027] According to the above scheme, in step S4, the farthest distance dis from all sampling points to the plane is calculated by combining the parameters related to the glass surface plane and the loosening. max The maximum difference in distance between adjacent data collection points and the plane (dif) max The maximum difference in the angle of the fitting plane normal vector of different groups of detection data, dia max Construct the feature set (dis) of the curtain wall glass. max dif max dia max ), and conduct loosening analysis.

[0028] According to the above scheme, in step S5, the neural network is a three-layer neural network, which consists of an input layer, a hidden layer and an output layer;

[0029] The input layer is responsible for receiving input from three features. Each input is processed by a neuron, and the activation function of the neuron is the ReLU function.

[0030] The hidden layer contains twenty neurons, which are responsible for receiving input from the input layer and performing nonlinear transformations. The neurons in the hidden layer are connected by weights and biases, and each connection has a ReLU activation function.

[0031] The output layer has only one neuron, which is responsible for converting the output of the hidden layer into the final output result. The activation function is the sigmoid function.

[0032] During training, the model uses the Adam optimizer to optimize parameters in order to minimize the loss function value.

[0033] The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations, as described in this invention, has the following beneficial effects:

[0034] 1. This invention can detect whether there are defects such as loosening while performing cleaning work on existing glass curtain walls of buildings. It obtains the scanning point information of the contact surface through coordinate changes and equipment parameter calculations, fits the plane of the existing glass curtain wall, extracts effective features from the point cloud information, and uses a pre-trained neural network to judge the defects of the glass curtain wall, so as to realize timely judgment and acquisition of glass curtain wall defects.

[0035] 2. This invention uses a contact detection module and a position detection module to work together to collect the coordinate information of the glass curtain wall during cleaning operations, enabling real-time detection of the looseness of the glass curtain wall and determining whether the glass needs to be replaced. This reduces the risk of glass falling or breaking. Compared with existing detection methods, this invention can shorten the detection cycle, reduce detection costs, and improve the safety of building glass curtain walls. Attached Figure Description

[0036] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0037] Figure 1 This is a schematic diagram of the structure of the intelligent detection and cleaning tool for detecting loose glass curtain walls according to the present invention;

[0038] Figure 2 This is a logic block diagram of the intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to the present invention;

[0039] Figure 3 This is a flowchart illustrating the intelligent detection method for loose glass curtain wall defects based on cleaning operations according to the present invention. Detailed Implementation

[0040] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0041] like Figure 1-3 As shown, the present invention discloses an intelligent detection device for loose glass curtain walls based on cleaning operations, comprising a glass curtain wall cleaning tool and a contact detection module, a pose detection module, a data processing, storage and transmission module and a data processing module disposed on the glass curtain wall cleaning tool; the glass curtain wall cleaning tool is a cleaning scraper.

[0042] The contact detection module measures the distance between the cleaning tool and the glass curtain wall, and uses the tool's structural parameters to design a threshold to detect whether the cleaning operation has begun. This module includes an ultrasonic distance sensor, which acquires the distance between the sensor on the device and the glass curtain wall based on ultrasonic ranging. The ultrasonic distance sensor is located on the lower outer side of the cleaning squeegee tip. The pose detection module obtains the device's pose information, thus determining the position of the cleaning contact point. This module is an IMU attitude sensor, including an accelerometer, angular velocity meter, and magnetometer, located on the upper outer side of the cleaning squeegee tip. The data processing, storage, and transmission module transmits pose detection data to the data processing module. This module is connected to the pose detection module via wires, enabling wired transmission of pose detection data. After processing and storage, the data is transmitted wirelessly via Bluetooth or the internet. The data processing module analyzes and processes the data collected by the contact and pose detection modules to achieve distributed detection of glass curtain wall loosening. The data processing, storage, and transmission module is located on the upper outer side of the cleaning scraper tip, behind the pose detection module.

[0043] This invention also provides an intelligent detection method for loosening defects in glass curtain walls based on cleaning operations. The method employs an intelligent detection device for loosening defects in glass curtain walls based on cleaning operations, and includes the following steps:

[0044] S1. Before carrying out cleaning operations, initialize the global coordinate system: take the three directions of northeast, south, and sky as the x, y, and z axes of the coordinate system, and take the position of the point at the start of the detection as the origin of the coordinate system.

[0045] The contact detection sensor detects the distance D between the device and the curtain wall glass. When the distance D is less than the set threshold [D], i.e., D < [D], the pose detection module sensor starts to collect data.

[0046] S2, with period τ i The system collects detection data from the contact detection module and the pose detection module, and then transmits the collected data to the data processing module through the data processing, storage and transmission module.

[0047] Instantaneous acceleration at corresponding sensor coordinates at each period time was collected using accelerometers, angular velocity meters, and magnetometers. instantaneous angular velocity and angles in the geomagnetic field coordinate system The angles are initialized by converting the obtained angles relative to the magnetic field into angles lower than those in the initial coordinate system, i.e.:

[0048] For any different time ti The sensor direction is below, and the predicted value is obtained by accumulating the instantaneous angular velocity gyro, that is: the t-th... i Instantaneous angular velocity in the coordinate system at time t Transform to the t-th node through rotation. i-1 angular velocity in the time coordinate system have in, For the tth i Time t i-1 The rotation coordinate transformation matrix at time t; for the t-th time... i-1 Time t i-2 At any time, there is Then for any time t i Its instantaneous acceleration in the initial coordinate system is Then at any time t i The angle prediction value is obtained by accumulating the instantaneous angular velocity gyro:

[0049] Considering the different error models between predicted and observed values, if the accumulated angular velocity is used as the predicted value and the angle obtained from the magnetic field is used as the observed value, then it is assumed that the sensor at any time t... i The angle below is Where, k pre and k obs For the weights of the predicted and observed values ​​relative to the true values, we have k. pre +k obs =1; Simultaneously, the actual value is used as a correction to the predicted value at that moment, resulting in...

[0050] For the change in displacement, at any time t i Below, there is Then there is Obtain the displacement vector in the initial coordinate system at each moment.

[0051] S3, the data processing module integrates the displacement vectors of each cycle. The sensor's detection center position in the initial coordinate system is obtained during the i-th cycle. for

[0052] Considering the sensor position and parameter settings at the contact point with the glass on the intelligent detection cleaning tool, i.e., in the sensor coordinate system, the coordinates of the two endpoints at the contact point are [x0 y0 z0] and [-x0 y0 z0], the line segment at the contact point is sampled into five points according to its length: [x0 y0 z0], [x0 / 2 y0 z0], [0 y0 z0], [-x0 / 2 y0 z0], and [-x0 y0 z0]. Combining this with the direction vector obtained in step S2, the corresponding time t on the plane in the initial coordinate system is obtained. i The coordinates of each collection point are shown below.

[0053] S4. The data processing module combines the location information of the collection points at all times and uses the least squares method to obtain the fitted glass surface plane.

[0054] Combining parameters related to the glass surface plane and loosening, calculate the farthest distance dis from all sampling points to the plane. max The maximum difference in distance between adjacent data collection points and the plane (dif) max The maximum difference in the angle of the fitting plane normal vector of different groups of detection data, dia max Construct the feature set (dis) of the curtain wall glass. max dif max dia max ), and conduct loosening analysis.

[0055] S5. The data processing module collects multiple sets of characteristic parameters corresponding to existing and loose glass and characteristic parameters corresponding to loose glass, establishes a three-layer fully connected neural network for classification, and after the neural network is established and supervised model training is completed, the neural network is deployed on the device to realize distributed detection of glass curtain wall loosening.

[0056] The neural network is a three-layer network consisting of an input layer, a hidden layer, and an output layer. The input layer receives three features, each processed by a neuron using the ReLU activation function. The hidden layer contains twenty neurons that receive input from the input layer and perform non-linear transformations. The neurons in the hidden layer are connected by weights and biases, with each connection having a ReLU activation function. The output layer has only one neuron, which transforms the output of the hidden layer into the final output, using the sigmoid activation function. During training, the model uses the Adam optimizer to optimize parameters and minimize the loss function value.

[0057] This invention proposes a cleaning tool for detecting loosening of existing glass curtain walls in buildings, comprising: a glass curtain wall cleaning tool, primarily a glass scraper; and an integrated curtain wall loosening detection device, integrated on the upper side of the cleaning tool, used to detect the loosening of the glass curtain wall during cleaning operations, enabling timely acquisition of glass curtain wall defects. The integrated curtain wall loosening detection device includes a contact detection module for calculating the distance between the cleaning tool and the glass curtain wall to detect whether the cleaning operation has commenced; and a pose detection module for obtaining the pose information of the device, thereby obtaining the position information of the cleaning contact point. The contact detection module includes an ultrasonic distance sensor. The pose detection module includes an accelerometer, angular velocity meter, and magnetometer integrated within an IMU attitude sensor module.

[0058] The main improvement of the intelligent detection method for loose glass curtain walls based on cleaning operations proposed in this invention is as follows: Based on an integrated detection device for cleaning operations, a contact detection module is equipped on the traditional curtain wall cleaning scraper. An ultrasonic distance sensor detects the distance between the cleaning scraper and the glass plane, thereby detecting whether the cleaning operation has begun. This device is also equipped with a posture detection module, which uses an accelerometer, angular velocity meter, and magnetometer to acquire information such as acceleration, angular velocity, and angle during operation, thereby obtaining the posture information of the device and the position information of the cleaning contact point.

[0059] The detection process of this invention is as follows:

[0060] The sensors on the detection equipment, including the contact detection module sensor and the pose detection module sensor, are activated, and the global coordinate system is initialized simultaneously. When the distance signal detected by the contact detection sensor is less than a set threshold, the pose detection module sensor begins to collect data. Data is collected at a set period. The pose detection module sensor collects data and performs noise reduction, collecting instantaneous acceleration and instantaneous angular velocity in the sensor coordinate system at corresponding time points of the period, as well as angles in the geomagnetic field coordinate system, and initializing the angles. The direction vector and displacement vector in the initial coordinate system are calculated at each time point. Combining the position information of the collected points at all time points, the fitted glass surface plane is obtained using the least squares method. The feature quantities of the point cloud with respect to the fitted plane are extracted, and after passing through a pre-trained neural network, distributed detection of glass curtain wall loosening is achieved, enabling timely judgment and acquisition of glass curtain wall defects.

[0061] The detection module uses an ultrasonic sensor to detect distance signals. When the detected distance is less than a set threshold, it is considered that the device has made contact with the glass surface, and the detection process begins. The pose detection module acquires and initializes the acceleration, angular velocity, and relative angle to the Earth's magnetic field from sensors at a set period. Combining the angular velocity and magnetometer detection models, and integrating the predicted value obtained by accumulating the angular velocity with the observed value detected by the magnetometer, a fitted true value is obtained. Using the sensor values ​​acquired at each moment, the direction vector and displacement vector at that moment are obtained. Combined with the device parameters, the coordinates of the acquisition point on the contact surface are calculated, the plane is fitted, and point cloud features are extracted. A pre-trained fully connected neural network is used to detect and classify the detected glass surface point cloud features to determine its looseness.

[0062] The present invention also provides a computer program for enabling a computer to execute a method for intelligent detection of loose glass curtain wall defects based on cleaning operations, thereby simplifying the implementation of the method for intelligent detection of loose glass curtain wall defects based on cleaning operations in any computing device.

[0063] The intelligent detection and cleaning tool for detecting loose glass curtain walls of the present invention includes a glass curtain wall cleaning tool, mainly a glass cleaning scraper; and an integrated detection device for loose glass curtain walls is integrated on the top of the cleaning tool and is used to detect the looseness of the glass curtain wall during cleaning operations, so as to realize the timely acquisition of glass curtain wall defects.

[0064] In the integrated glass curtain wall loosening detection equipment, the ultrasonic distance sensor for contact detection is located on the lower outer side of the cleaning squeegee tip, used to detect whether the squeegee is in contact with the curtain wall glass during cleaning operations. The pose detection module for pose detection is located on the upper outer side of the cleaning squeegee tip, used to collect pose data of the cleaning equipment during cleaning operations. The data processing, storage, and transmission module is located on the upper outer side of the cleaning squeegee tip, behind the pose detection module. The data processing, storage, and transmission module is connected to the pose detection module via wires, used to transmit pose detection data to the data processing unit via wired connection. After data processing and storage are completed, the data is transmitted wirelessly via Bluetooth or the Internet.

[0065] Example

[0066] A method for calculating the position of a cleaning tool during movement includes the following steps:

[0067] S1. Before starting the cleaning operation, activate the sensors on the equipment, including the contact detection module sensor and the pose detection module sensor, and initialize the global coordinate system. When the distance signal detected by the ultrasonic sensor of contact detection is less than the set threshold, i.e., D < [D] = 25mm, the pose detection module sensor begins to collect data. v0 is the speed of sound in the environment, set to v0 = 340 m / s; Δt is the time interval between emitting and receiving the ultrasonic signal; [D] is the contact detection threshold set according to the tool's structural parameters, set as [D] = 25 mm = 2.5 × 10⁻⁶. -2 m.

[0068] S2, with period τ i Data acquisition begins; the pose detection module sensor starts collecting data and then performs smoothing and noise reduction. The denoising result is as follows:

[0069] Data collection results

[0070]

[0071] Instantaneous acceleration at corresponding sensor coordinates at each period time was collected using accelerometers, angular velocity meters, and magnetometers. instantaneous angular velocity and angles in the geomagnetic field coordinate system The angles are initialized by converting the obtained angles relative to the magnetic field into angles lower than those in the initial coordinate system, i.e.: The angle was observed using a magnetometer.

[0072] For any different time t i The sensor direction can also be predicted by accumulating the instantaneous angular velocity gyro, that is: the t-th... i Instantaneous angular velocity in the coordinate system at time t Transform to the t-th node through rotation. i-1 angular velocity in the time coordinate system have in For the tth i Time t i-1 The rotation coordinate transformation matrix at time t. i-1 Time t i-2 At any time, there is Then for any time t i Its instantaneous acceleration in the initial coordinate system is Then at any time t i The angle prediction value is obtained by accumulating the instantaneous angular velocity gyro:

[0073] Considering the different error models between predicted and observed values, if the accumulated angular velocity is used as the predicted value and the angle obtained from the magnetic field is used as the observed value, then it is assumed that the sensor at any time t... i The angle below is Where k pre and k obsThe weights of the predicted and observed values ​​relative to the true values ​​are obtained from the detection error model analysis of both, k. pre =0.35, k obs =0.65. The detection error models for both are normally distributed. There is k pre +k obs =1.

[0074] At the same time, the actual value is used as a correction to the predicted value at the next time step.

[0075] Note that gravitational acceleration is involved in the measurement process, therefore the influence of gravity at different angles is eliminated, i.e.

[0076] For the change in displacement, at any time t i Below, there is Then there is Obtain the displacement vector in the initial coordinate system at each moment.

[0077] R x R y R z These are the rotation transformation matrices around the x, y, and z axes, respectively.

[0078]

[0079]

[0080]

[0081] Rotate around the zyx axis sequentially, with the rotation matrix as follows:

[0082]

[0083] In actual cleaning operations, due to the small sampling period and the slow equipment rotation speed caused by manual labor, the trigonometric measurements of α, β, and γ in the rotation matrix are close to 0 in each coordinate change. Therefore, a small angle approximation is used for the above matrix, namely: sin(α)≈α, cos(α)≈1.

[0084] Therefore, the matrix can be calculated equivalently:

[0085]

[0086] The above matrix represents rotations along the z, y, and x axes in a specific order. Considering the impact of different orders, matrix calculations are performed in different orders, but the final approximate rotation matrix is ​​the same. Therefore, the displacement rotation is calculated using the above rotation matrix T.

[0087] After sampling for each cleaning operation, the displacement in the local coordinate system under the last sampling period is iteratively transferred to the coordinate system of the previous moment, finally obtaining the displacement under each sampling period in the initial coordinate system. By continuously accumulating these displacements, the displacement at any moment t during cleaning in the initial coordinate system is obtained. i The pose information of the device.

[0088] S3. By integrating the displacement vectors at each time point, we can obtain the displacement vector at any time t. i Below, the sensor detection center position is located in the initial coordinate system as follows:

[0089] Obtain the sensor detection center position in the example

[0090]

[0091] Considering the sensor position and parameter settings at the contact point with the glass on the intelligent detection cleaning tool, i.e., the coordinates of the two endpoints at the contact point are [x0y0z0] and [-x0y0z0] in the sensor coordinate system, the line segment at the contact point is sampled into five points based on its length: [x0y0z0], [x0 / 2y0z0], [0y0z0], [-x0 / 2y0z0], and [-x0y0z0]. Combining this with the direction vector obtained in step S2, the corresponding time t on the plane in the initial coordinate system is obtained. i The coordinates of each data acquisition point are shown below. x0 = 211mm, y0 = 42mm, and z0 = 5mm are obtained based on the structural parameters.

[0092] By combining the location information of the data collection points at all time points, the fitted glass surface plane is obtained using the least squares method. Then, by incorporating its loosening-related parameters, the farthest distance *dis* between all data collection points and the plane is calculated. max The maximum difference in distance between adjacent data collection points and the plane (dif) max The maximum difference between the angle and the perpendicular angle between different contact line vectors and the fitting plane normal vector is dia. max Construct the feature set (dis) of this curtain wall glass. max dif max dia max ), and conduct loosening analysis.

[0093] In a preferred embodiment of the present invention, the equation of the fitted plane is obtained as Ax + By + Cz + D = 0, and the equation of the fitted plane at time i is (x + By + Cz + D = 0). i,j y i,j z i,j The distance to the fitted plane is

[0094] In a preferred embodiment of the present invention, the feature set (dis) in the detectionmax dif max dia max (0.0031 0.0013 7.31°).

[0095] S5. By collecting multiple sets of characteristic parameters corresponding to existing and loose glass and characteristic parameters corresponding to loose glass, a neural network is established for classification. After the neural network is established and supervised model training is completed, the neural network is deployed on the device to realize distributed detection of glass curtain wall loosening.

[0096] By collecting multiple sets of feature parameters corresponding to both existing and loose glass, and feature parameters corresponding to loose glass, a neural network is established for classification. This neural network model is a three-layer neural network, consisting of an input layer, a hidden layer, and an output layer.

[0097] Input layer: This layer receives input from three features. Each input is processed by a neuron, with the neuron's activation function being the ReLU (Modified Linear Unit) function.

[0098] Hidden layer: This layer contains twenty neurons responsible for receiving input from the input layer and performing non-linear transformations. The neurons in the hidden layer are connected by weights and biases, and each connection has a ReLU activation function.

[0099] Output layer: This layer has only one neuron and is responsible for transforming the output of the hidden layer into the final output. The activation function used is the sigmoid function.

[0100] During training, the model uses the Adam optimizer to optimize parameters and minimize the loss function value. For binary classification problems, the loss function is binary cross-entropy, and the evaluation metric is accuracy. After building the neural network and completing supervised model training, the neural network is deployed on a device to achieve distributed monitoring of glass curtain wall loosening.

[0101] In a preferred embodiment of the present invention, the feature set (dis) in the detection max dif max dia max After classification using a neural network, the glass falls into the "no loosening" category, indicating that the curtain wall glass did not exhibit any loosening during this inspection. The collected data on glass loosening is stored and transmitted to a platform via Bluetooth or the internet. This platform enables visualized operation of the entire building's glass curtain wall, facilitating subsequent operation and maintenance management.

[0102] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for intelligent detection of loosening defects in glass curtain walls based on cleaning operations, characterized in that, An intelligent detection device for loose glass curtain walls based on cleaning operations is adopted. The intelligent detection device includes glass curtain wall cleaning tools and contact detection module, posture detection module, data processing, storage and transmission module and data processing module installed on the glass curtain wall cleaning tools. The contact detection module is used to calculate the distance between the cleaning tool and the glass curtain wall, and to design a threshold based on the structural parameters of the cleaning tool to detect whether the cleaning operation has started. The pose detection module is used to obtain the pose information of the device, thereby obtaining the position information of the cleaning contact point; The data processing, storage, and transmission module is used to transmit pose detection data to the data processing module. The data processing module is used to analyze and process the data collected by the contact detection module and the pose detection module to realize distributed detection of glass curtain wall loosening. The detection method includes the following steps: S1. Before starting the cleaning operation, initialize the global coordinate system: using the three directions of northeast, sky, and zenith as the coordinate system. , , The coordinate origin is set along the axis, with the position of the point at the start of the detection. S2, in a periodic manner The system collects detection data from the contact detection module and the pose detection module, and then transmits the collected data to the data processing module through the data processing, storage and transmission module. S3, the data processing module integrates the displacement vectors of each cycle. , obtained in the first In each cycle, the sensor detects the center position in the initial coordinate system. for ; S4. The data processing module combines the location information of the collection points at all times and uses the least squares method to obtain the fitted glass surface plane. S5. The data processing module collects multiple sets of characteristic parameters corresponding to existing and loose glass and characteristic parameters corresponding to loose glass, establishes a three-layer fully connected neural network for classification, and after the neural network is established and supervised model training is completed, the neural network is deployed on the device to realize distributed detection of glass curtain wall loosening.

2. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, In step S1, the distance between the contact detection sensor and the curtain wall glass is measured. Distance detected Less than the set threshold At that time, there was The pose detection module sensor begins collecting data.

3. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, In step S2, the instantaneous acceleration at the sensor coordinates at corresponding time points in the cycle is collected by the accelerometer, angular velocity meter, and magnetometer. Instantaneous angular velocity and angles in the geomagnetic field coordinate system First, initialize the angle by converting the obtained angle relative to the magnetic field into an angle lower than the initial coordinate system, i.e.: ; For any different time The sensor direction is downward, measured by instantaneous angular velocity. The predicted value is obtained by accumulating the values, that is: the first... Instantaneous angular velocity in the coordinate system at time t Transformed to the th through rotation angular velocity in the time coordinate system ,have ,in, For the first Time to the The rotation coordinate transformation matrix at time t; for the t... Time to the At any time, there is , Then for any given moment... Its instantaneous acceleration in the initial coordinate system is Then at any given moment Through instantaneous angular velocity The cumulative angle prediction value is: ; Considering the different error models between predicted and observed values, if the accumulated angular velocity is used as the predicted value and the angle obtained from the magnetic field is used as the observed value, then it is assumed that the sensor at any given moment... The angle below is ,in, and For the weights of the predicted and observed values ​​relative to the true values, we have Simultaneously, the actual value is used as a correction to the predicted value at that moment. ; For the change in displacement, at any given moment Below, there is , Then there is This yields the displacement vector in the initial coordinate system at each moment.

4. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, In step S3, the parameter settings for the sensor position and the contact point with the glass on the intelligent detection cleaning tool are considered; that is, in the sensor coordinate system, the coordinates of the two endpoints at the contact point are... and The line segment at the contact point is sampled into five points based on its length, i.e. , , , , Combining the direction vector obtained in step S2, we obtain the corresponding time on the plane in the initial coordinate system. The coordinates of each collection point are shown below.

5. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, In step S4, the farthest distance between all sampling points and the plane is calculated by combining the parameters related to the glass surface plane and loosening. The maximum difference in distance between adjacent data collection points and the plane. The maximum difference in the angle of the fitting plane normal vector of different groups of test data Constructing the feature set of curtain wall glass We will conduct a loosening analysis.

6. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, In step S5, the neural network is a three-layer neural network, which consists of an input layer, a hidden layer and an output layer; The input layer is responsible for receiving input from three features. Each input is processed by a neuron, and the activation function of the neuron is the ReLU function. The hidden layer contains twenty neurons, which are responsible for receiving input from the input layer and performing nonlinear transformations. The neurons in the hidden layer are connected by weights and biases, and each connection has a ReLU activation function. The output layer has only one neuron, which is responsible for converting the output of the hidden layer into the final output result. The activation function is the sigmoid function. During training, the model uses the Adam optimizer to optimize parameters in order to minimize the loss function value.

7. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, The contact detection module includes an ultrasonic distance sensor, which acquires the distance between the sensor on the device and the curtain wall glass, obtained based on ultrasonic ranging.

8. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, The pose detection module is an IMU attitude sensor, which includes an accelerometer, an angular velocity meter, and a magnetometer.

9. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, The data processing, storage, and transmission module is connected to the pose detection module via a wire. The data processing, storage, and transmission module is used to transmit pose detection data to the data processing module via wire. After completing data processing and storage, the data is sent wirelessly via Bluetooth or the Internet.

10. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 1, characterized in that, The glass curtain wall cleaning tool is a cleaning scraper.

11. The intelligent detection method for loosening defects in glass curtain walls based on cleaning operations according to claim 7, characterized in that, The ultrasonic distance sensor is located on the lower outer side of the cleaning squeegee tip, the pose detection module is located on the upper outer side of the cleaning squeegee tip, and the data processing, storage, and transmission module is located on the upper outer side of the cleaning squeegee tip, behind the pose detection module.

Images