Door body damage detection method and device, electronic equipment and storage medium

Abstract

The invention provides a door body damage detection method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining a door body damage recognition agent through reinforcement learning of a door body damage detection model; identifying the door opening angle sequence and the door opening force sequence of the to-be-detected door body in a preset time sequence by using a door body damage identification agent, and outputting a damage coefficient sequence of each door body core component of the to-be-detected door body in the preset time sequence; and determining a detection result of the to-be-detected door body based on the damage coefficient sequence of each door body core component. According to the method, after a 3D parameterized damage behavior simulation model considering damage is established, order reduction is performed on a discrete damage behavior simulation model through deep learning to obtain a door body damage detection model, and then the door body damage detection model considering damage evolution behaviors is learned based on a reinforcement learning algorithm to obtain a door body damage identification agent; and real-time monitoring of the health state of the door body system can be realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a door damage detection method, device, electronic equipment and storage medium. Background Art

[0002] The door system of appliances such as dishwashers may have core component damage during long-term use, affecting the performance and safety of the equipment. Therefore, real-time and accurate monitoring of the damage status of the door system to improve the reliability and service life of the equipment is a demand of the current industry.

[0003] In the traditional door system health management mode, there are generally two ways to obtain the health status of the door system: 1) Collect customer feedback. During the use of the door system, if the customer finds that the door system is abnormal, he will contact the customer service for feedback. The after-sales team will investigate the abnormal situation and find the cause of the failure of the door system. Through long-term accumulation, relevant failure data can be collected. 2) By establishing a corresponding laboratory, the door system fast closing accelerated fatigue endurance life test is carried out. This method can be used to understand the evolution of the health status of the door system, but due to the differences in the timing of the actions of the door system during actual use, the damage evolution results are inaccurate.

[0004] It can be seen that there is currently no technology in the industry that can monitor damage to door systems in real time, and there is an urgent need to fill this gap in the industry. Summary of the invention

[0005] The present invention provides a door damage detection method, device, electronic device and storage medium, which are used to solve the defect that the current detection of the door system of household appliances cannot achieve real-time and accurate detection.

[0006] The present invention provides a door body damage detection method, comprising the following steps: Through reinforcement learning of the door damage detection model, the door damage recognition agent is obtained; The door body damage identification intelligent agent is used to identify the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; Based on the damage coefficient sequence of each door body core component, the detection result of the door body to be tested is determined.

[0007] According to a door body damage detection method provided by the present invention, the door body damage detection model is constructed by adopting the following steps: Based on the physical benchmarking data and the damage coefficient of each door body core component at each discrete time point in the preset time sequence, a damage behavior simulation model of each discrete time point is constructed; the physical benchmarking data is obtained by performing physical opening and closing experiments on each door body core component of a real physical door body; The first initial network model is trained using a training data set to obtain the door body damage detection model; any training data in the training data set includes a door opening angle of the damage behavior simulation model at its corresponding discrete time point and the damage coefficient of each door body core component, and the label of the training data is the door opening force at the discrete time point.

[0008] According to a door body damage detection method provided by the present invention, the damage behavior simulation model of each discrete time point is constructed based on the physical benchmarking data and the damage coefficient of each door body core component at each discrete time point in a preset time sequence, including: Using the physical benchmarking data to benchmark each non-destructive simulation model one by one, to obtain a normal behavior simulation model of each of the door body core components; The normal behavior simulation models are assembled to form a door system simulation model; According to the damage coefficient of each of the door body core components at each discrete time point on the preset time sequence, the door body system simulation model is updated to obtain the damage behavior simulation model under different damage states at each discrete time point.

[0009] According to a door body damage detection method provided by the present invention, the physical benchmarking data includes force time series curve samples obtained by performing physical opening and closing experiments under angle control on each of the door body core components; The method of using the physical benchmarking data to benchmark each lossless simulation model one by one to obtain a normal behavior simulation model of each door body core component includes: For the lossless simulation model of any core component of the door body, the model parameters of the lossless simulation model are adjusted according to the curve difference between the virtual force timing curve output by the lossless simulation model and the force timing curve sample until the curve difference is within a preset range, and the adjusted lossless simulation model is obtained as the normal behavior simulation model.

[0010] According to a door body damage detection method provided by the present invention, the door body core component includes a complete door body system and all key subsystems, and the key subsystem includes one or more of a hinge subsystem, a pull rope spring subsystem, a lock subsystem and a sealing subsystem; the hinge subsystem is composed of a skeleton subsystem and an outer door subsystem; Correspondingly, the non-destructive simulation model includes a complete door body system simulation model, a hinge subsystem simulation model, a pull rope spring subsystem simulation model, a lock buckle subsystem simulation model and a sealing subsystem simulation model.

[0011] According to a door body damage detection method provided by the present invention, the normal behavior simulation model is assembled to form a door body system simulation model, including: The normal behavior simulation models of the complete door body system, hinge subsystem, pull cord spring subsystem, lock buckle subsystem and sealing subsystem corresponding to the complete door body system simulation model, the hinge subsystem simulation model, the pull cord spring subsystem simulation model, the lock buckle subsystem simulation model and the sealing subsystem simulation model are integrated to obtain the door body system simulation model.

[0012] According to a door body damage detection method provided by the present invention, the door body system simulation model is updated according to the damage coefficient of each door body core component at each discrete time point in the preset time sequence, and the damage behavior simulation model under different damage states at each discrete time point is obtained, including: Determine the damage coefficient of any door body core component at any discrete time point based on the current opening and closing times of any door body core component at any discrete time point and the fatigue opening and closing times limit of any door body core component; Determine a scaling factor of a target parameter of any door body core component related to door body damage based on the damage coefficient; At any discrete time point, the door system simulation model is updated according to the scaling coefficients of target parameters of all door core components related to door damage to obtain the damage behavior simulation model at any discrete time point; All the discrete time points are traversed to obtain all the damage behavior simulation models.

[0013] According to a door body damage detection method provided by the present invention, the target parameters of the hinge subsystem include the deformation parameters at the connection point between the hinge and the pull rope, the target parameters of the pull rope spring subsystem include the pull rope stiffness and the telescopic spring stiffness, the target parameters of the lock subsystem include the locking spring stiffness, and the target parameters of the sealing subsystem include the sealing strip rebound force.

[0014] According to a door damage detection method provided by the present invention, a first initial network model is trained using any training data in a training data set to obtain the door damage detection model, including: According to the preset time sequence, the training data corresponding to each discrete time point is sequentially input into the first initial network model to obtain a door opening force prediction sequence output by the first initial network model; According to the network error between the door opening force prediction sequence and the door opening force label sequence, the model parameters of the first initial network model are gradually updated until the model converges, thereby obtaining the door body damage detection model; The door opening force label sequence is composed of the door opening forces at all discrete time points on the preset time sequence.

[0015] According to a door damage detection method provided by the present invention, the door damage recognition agent is obtained by reinforcement learning of a door damage detection model, including: A reinforced training set is constructed using the door body damage detection model; any reinforced training data in the reinforced training set includes a door opening angle and a door opening force at any discrete time point, and the label of any reinforced training data is a damage coefficient of each door body core component at any discrete time point, and the door opening force is obtained by inputting the door opening angle and the damage coefficient of each door body core component into the door body damage detection model; The second initial network model is trained using the enhanced training set to obtain the door damage recognition intelligent agent.

[0016] According to a door damage detection method provided by the present invention, the second initial network model is trained using the enhanced training set to obtain the door damage recognition agent, including: According to the preset time sequence, the enhanced training data corresponding to each discrete time point is sequentially input into the second initial network model to obtain a damage coefficient prediction sequence output by the second initial network model; According to the network error between the damage coefficient prediction sequence and the damage coefficient label sequence, the model parameters of the second initial network model are gradually updated until the model converges, thereby obtaining the door body damage recognition intelligent agent; The damage coefficient label sequence is composed of the damage coefficients of each door core component at all discrete time points on the preset time sequence.

[0017] According to a door damage detection method provided by the present invention, the first initial network model and the second initial network model are both obtained based on the long short-term memory network model as the basic model architecture.

[0018] The present invention also provides a door damage detection device, comprising: An agent training unit, used for obtaining a door damage recognition agent through reinforcement learning of a door damage detection model; An intelligent agent recognition unit is used to use the door body damage recognition intelligent agent to recognize the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; The door body detection unit is used to determine the detection result of the door body to be detected based on the damage coefficient sequence of each door body core component.

[0019] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a door damage detection method as described above is implemented.

[0020] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the door damage detection method as described in any one of the above is implemented.

[0021] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the door damage detection method as described above is implemented.

[0022] The door body damage detection method, device, electronic device and storage medium provided by the present invention, after establishing a 3D parameterized damage behavior simulation model considering damage, reduce the order of the discrete damage behavior simulation model through deep learning to obtain a door body damage detection model, and then learn the door body damage detection model considering the damage evolution behavior based on the reinforcement learning algorithm to obtain a door body damage recognition intelligent agent, which can realize real-time monitoring of the health status of the door system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 It is one of the flow charts of the door damage detection method provided by the present invention.

[0025] Figure 2 This is the second flow chart of the door damage detection method provided by the present invention.

[0026] Figure 3 It is a flow chart of the method for constructing the door damage detection model provided by the present invention.

[0027] Figure 4 It is a schematic diagram of the hinge subsystem damage coefficient determination method provided by the present invention.

[0028] Figure 5 It is a schematic diagram of the stiffness curve of the sealing strip provided by the present invention.

[0029] Figure 6 It is a schematic diagram of the training process of the door damage detection model provided by the present invention.

[0030] Figure 7 It is a structural schematic diagram of the door damage detection device provided by the present invention.

[0031] Figure 8 This is one of the schematic diagrams showing the performance of the door damage detection model provided by the present invention in the training data set.

[0032] Fig. 9 This is the second schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set.

[0033] Fig.10 This is the third schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set.

[0034] Fig.11 This is the fourth schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set.

[0035] Fig.12 This is the fifth schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set.

[0036] Fig.13 This is one of the schematic diagrams showing the performance of the door damage detection model provided by the present invention in the test data set.

[0037] Fig.14 This is the second schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set.

[0038] Fig.15 This is the third schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set.

[0039] Fig.16 This is the fourth schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set.

[0040] Fig.17 This is the fifth schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set.

[0041] Fig.18 It is one of the schematic diagrams of the evolution of the loss function value of the door damage detection model provided by the present invention during the training process.

[0042] Fig.19 This is the second schematic diagram of the evolution of the loss function value of the door damage detection model provided by the present invention during the training process.

[0043] Fig. 20It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be noted that in the description of the present invention, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. The orientation or position relationship indicated by the terms "upper", "lower" and the like is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0046] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0047] The health status management scenarios of electrical equipment, especially household electrical equipment, is a very broad category. Taking the dishwasher as an example, in the actual use of the user, the scenarios that require health status management are: 1) Monitoring the health status of the dishwasher washing system: During the washing process, whether the water pressure is abnormal, whether the water temperature is abnormal, whether the booster pump is abnormal, whether the respirator is abnormal, whether the drainage system is abnormal, etc. The monitoring and prediction of the health status of the entire washing system, and fault diagnosis are generally related to the relevant control system; 2) Monitoring the health status of the dishwasher structure system: Whether the various systems of the dishwasher, including the door system, the inner tank system, the chassis system, etc., have been damaged, analyze the damage value, the remaining life, and determine whether the life limit has been reached, etc.

[0048] During user use, the door system of electrical equipment, such as the pull rope spring subsystem, sealing strip subsystem, lock buckle subsystem, hinge subsystem and other core components of the door body, are likely to malfunction after a certain number of years of use. For example, the following malfunctions may occur: 1) Pull rope spring subsystem: the telescopic spring becomes loose or broken, and the pull rope becomes worn or broken; 2) Sealing strip subsystem: the sealing performance of the sealing strip gradually decreases, and the rebound force decreases; 3) Lock buckle subsystem: the pin position of the lock buckle becomes worn, and the locking spring becomes loose or broken; 4) Hinge subsystem: the hinge structure becomes deformed or broken, etc.

[0049] During the design process of the door system, in order to quickly obtain the damage evolution process of the door system throughout its life cycle, a corresponding laboratory will be established to conduct fast-closing accelerated fatigue endurance life tests on the door system. At the same time, the fault data obtained by fault analysis will be combined with the feedback from customers during use. However, since the action sequence of each door system is complex and personalized, the damage evolution may be fast or slow. Therefore, these existing approaches often cannot achieve systematic and real-time analysis of door system damage. How to solve this demand is an urgent problem that needs to be solved in the industry.

[0050] In today's rapidly developing digital world, the present invention provides a door body damage detection method, device, electronic device and storage medium based on the integration of digital twin technology, reinforcement learning algorithm and deep learning algorithm, in order to provide a new health management mode for the door body system.

[0051] The present invention proposes a door damage detection method, device, electronic device and storage medium, and its main implementation steps include the following steps: First, design a model test to obtain physical benchmarking data as the benchmarking standard for the non-destructive simulation model. Secondly, construct a normal behavior simulation model by benchmarking the physical benchmarking data. Thirdly, based on the modified normal behavior simulation model, fatigue damage theory is used to propose reasonable mechanical assumptions, establish a damage behavior simulation model that considers damage evolution, and output a training data set for deep learning. Then, based on the deep learning network algorithm, training is performed to output the corresponding AI reduced-order model of the damage behavior simulation model as a high-fidelity digital twin model that considers damage evolution behavior, namely, the door body damage detection model. Subsequently, the door body damage detection model can be learned based on the reinforcement learning algorithm to obtain the corresponding intelligent agent network model, which can be used for real-time monitoring of the damage status of the door body system.

[0052] Combine the following Figure 1-Figure 20 The present invention describes a door damage detection method, device, electronic device and storage medium.

[0053] Figure 1 FIG. 1 is one of the flow charts of the door damage detection method provided by the present invention, such as Figure 1 As shown, including but not limited to the following steps: Step 101, obtaining a door damage recognition agent through reinforcement learning of a door damage detection model.

[0054] First, we need to build a high-fidelity door damage detection model. This door damage detection model is built based on digital twin technology and deep learning algorithms. It can simulate the behavior of the door system under different damage states. The specific construction steps mainly include but are not limited to: Physical opening and closing tests are conducted on the core components of the real physical door body (such as the hinge subsystem, rope spring subsystem, lock subsystem and sealing subsystem) to obtain physical benchmarking data. Angle control can be used in the test to record the door opening force data at different opening and closing angles to form a force time series curve sample.

[0055] Based on the physical benchmarking data, each lossless simulation model is benchmarked to obtain the normal behavior simulation model of each door core component. These normal behavior simulation models are combined to form a complete door system simulation model.

[0056] According to the fatigue characteristics of each core component of the door body, the damage coefficient is defined, and the relevant parameters in the door body system simulation model are updated based on the damage coefficient to generate a damage behavior simulation model under different damage states.

[0057] The initial network model (such as long short-term memory network LSTM) is trained using the training data set to obtain the door damage detection model. The training data includes the door opening angle, damage coefficient, and corresponding door opening force label of the damage behavior simulation model at each discrete time point.

[0058] Finally, the door damage detection model is optimized through the reinforcement learning algorithm to obtain the door damage recognition agent. The specific implementation steps can be described as follows: The behavior simulation environment of the door system is defined as a reinforcement learning environment, where the state is the sequence of door opening angles and damage coefficients, and the action is the decision variable in space (such as the change of door opening force).

[0059] The reward function is designed based on the closeness between the door-opening force output by the model and the actual damage state. For example, when the error between the output door-opening force sequence and the actual door-opening force sequence is small, a positive reward is given; otherwise, a negative reward is given.

[0060] The door damage detection model is trained using a reinforcement learning algorithm (such as Q-learning, DQN or PPO) so that the agent can learn the optimal strategy in a given environment and output an accurate damage coefficient sequence.

[0061] Step 102, using the door body damage identification intelligent agent to identify the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence.

[0062] In practical applications, when performing damage detection on the door body to be tested, it is necessary to collect the door opening angle sequence and door opening force sequence within the preset time sequence. The specific method is as follows: High-precision force sensors and angle sensors are installed on the door to be tested to collect door opening force and door opening angle data in real time.

[0063] Angle control is used to ensure that the angle increment of each door opening is consistent (for example, 1° each time), and the door opening force and door opening angle at each discrete time point are recorded.

[0064] The collected door opening angle sequence and door opening force sequence are input into the door damage recognition agent trained by reinforcement learning. The door damage recognition agent outputs the damage coefficient sequence of each core component of the door body to be tested within the preset time sequence according to the learned strategy.

[0065] Step 103: determining the detection result of the door body to be tested based on the damage coefficient sequence of each door body core component.

[0066] The damage degree of each door core component can be evaluated based on the damage coefficient sequence. For example, when the damage coefficient of a door core component is close to 0, it means that the door core component is in good condition; when the damage coefficient of a door core component is close to 1, it means that the door core component is close to the service life limit.

[0067] Finally, the damage coefficient sequence of each core component of the door body can be combined to determine the overall health status of the door body to be tested. The specific method is as follows: if the damage coefficient of at least one core component of the door body is high, it is judged that the overall damage of the door body to be tested is serious and needs maintenance or replacement; if the damage coefficient of most core components is low, it is judged that the door body to be tested is in good condition and can continue to be used.

[0068] Figure 2 This is the second flow chart of the door damage detection method provided by the present invention. Figure 2 As shown, the detection of the dishwasher door system is taken as an example for explanation.

[0069] Taking the dishwasher door system as an example, its core components include hinge subsystem, rope spring subsystem, lock subsystem and sealing subsystem, etc. Through the above method, real-time damage detection of the dishwasher door system can be achieved. The specific steps are as follows: First, the normal behavior simulation model of each door core component in the dishwasher door system is constructed, and all normal behavior simulation models are combined to form a door system simulation model. Then, the damage evolution mechanism is introduced to generate damage behavior simulation models under different damage states, so as to use the damage behavior simulation model to construct a training data set for model training of the initial network model.

[0070] The training data corresponding to each discrete time point in the training data set is input into the initial network model in turn, including recording the input data and label data of each discrete time point from the training data set, and inputting the input data (door opening angle and damage coefficient) into the initial network model.

[0071] Run the initial network model and obtain the door opening force prediction sequence output by it, that is, use the initial network model to infer the input data and generate the door opening force prediction sequence (collect the prediction results of each discrete time point to form a complete door opening force prediction sequence).

[0072] The network error between the prediction sequence and the label sequence is calculated using the mean square error (MSE) or other statistical indicators. The parameters of the initial network model can be gradually updated according to the calculated network error until the model converges to obtain a door damage detection model. The door damage detection model can accurately predict the corresponding door opening force according to the input door opening angle and damage coefficient.

[0073] Finally, a reinforcement training set is constructed through the door damage detection model (the door opening angle sequence and door opening force sequence simulated by the door damage detection model are taken as input, and the corresponding damage coefficient is used as a label to construct a reinforcement learning training set). The generated reinforcement learning training set is used to train a new initial network model to obtain the required door damage recognition intelligent agent.

[0074] By using the door body damage recognition intelligent agent to analyze the real-time collected door opening force and door opening angle data, the damage coefficient sequence of each door body core component within the preset time sequence can be obtained, so as to evaluate the damage degree of each door body core component based on the damage coefficient sequence of each door body core component. By combining the damage coefficient sequence of each door body core component, the overall health status of the door body to be tested can be determined quickly and accurately.

[0075] The door body damage detection method provided by the present invention, after establishing a 3D parameterized damage behavior simulation model taking damage into account, reduces the order of the discrete damage behavior simulation model through deep learning to obtain a door body damage detection model, and then learns the door body damage detection model taking into account the damage evolution behavior based on the reinforcement learning algorithm to obtain a door body damage recognition intelligent agent, which can realize real-time monitoring of the health status of the door body system.

[0076] The following is a detailed introduction to the method for constructing the door damage detection model provided by the present invention.

[0077] Figure 3 FIG. 1 is a flow chart of a method for constructing a door damage detection model provided by the present invention. Figure 3 As shown, including but not limited to the following steps: Step 301 : constructing a damage behavior simulation model at each discrete time point based on physical benchmarking data and the damage coefficient of each door body core component at each discrete time point in a preset time sequence.

[0078] The physical benchmarking data is obtained by performing physical opening and closing experiments on the core components of a real physical door body, and the specific acquisition method thereof may be: Select a representative dishwasher door system to ensure that its core components (such as the rope spring subsystem, sealing subsystem, lock subsystem, etc.) are in the initial undamaged state. In the physical opening and closing experiment, the angle control method is adopted, that is, the angle increment of each door opening is kept consistent (for example, 1° each time), and the door opening force data of the door system at different opening and closing angles are recorded. Specifically, high-precision force sensors and angle sensors can be used to collect door opening force data and door opening angles at each discrete time point (i.e., discrete sampling time point) of the preset time sequence to form door opening force timing data and door opening angle timing data. The obtained door opening force timing data and door opening angle timing data can be represented by the door opening force timing curve and the door opening angle timing curve, respectively. These data will be used as physical benchmarking data for subsequent benchmarking of the non-destructive simulation model.

[0079] The damage coefficient is mainly used to quantify the degree of damage of each door core component at different stages of use. It can generally be determined by the ratio between the current opening and closing times of each door core component and the fatigue opening and closing times limit.

[0080] For example, for the cable spring subsystem, the damage coefficient D at any discrete time point can be defined as: ,in is the current opening and closing times at any discrete time point, This is the limit value of the fatigue opening and closing times of the rope-spring subsystem.

[0081] Furthermore, based on the physical benchmarking data and damage coefficients, a damage behavior simulation model for each discrete time point can be constructed, which can be achieved by taking the following steps: Firstly, according to the initial parameters of each door body core component, an initialization simulation model of each door body core component in the door body system is constructed, and the initialization simulation model can be called a lossless simulation model.

[0082] Then, by using the physical benchmarking data to benchmark each lossless simulation model, the normal behavior simulation model of each door core component can be obtained. The door system simulation model can be constructed using the normal behavior simulation model of all door core components.

[0083] Finally, at each discrete time point on the preset time sequence, the parameters of the corresponding components in the door system simulation model are updated according to the loss coefficient of each door core component at the discrete time point. For example, at a discrete time point, the stiffness of the rope spring is adjusted according to the damage coefficient D of the rope spring subsystem. In this way, at each discrete time point, a damage behavior simulation model that takes into account the damage evolution behavior can be obtained.

[0084] Assuming that 50 discrete sampling points are set on the entire preset time sequence, 50 damage behavior simulation models under different damage states can be obtained.

[0085] Step 302: train the initial network model using the training data set to obtain the door damage detection model.

[0086] Among them, any training data in the training data set includes a door opening angle of the damage behavior simulation model at its corresponding discrete time point and the damage coefficient of each door body core component, and the label of the training data is the door opening force at the discrete time point.

[0087] Specifically, a training data set can be prepared first, and each training data in the training data set is closely related to the damage behavior simulation model obtained at each discrete sampling point. First, the door opening force data of the door system at different opening and closing angles within the preset time sequence is tested by experiment, so that the door opening angle and door opening force of each discrete sampling point can be obtained. At the same time, according to the opening and closing times of each door core component in the door system at each discrete sampling point, the damage coefficient of each door core component is counted.

[0088] Furthermore, the door opening angle obtained at each discrete time point and the damage coefficient of each door core component are combined into a training data, and the door opening force obtained at the discrete time point is used as the label of the training data, so that a set of training samples corresponding to each discrete time point can be obtained. Assume that there are five core components of the door body, and the damage coefficients at any discrete time point are D1, D2, D3, D4 and D5 respectively. Then the input of the training sample at any discrete time point = (door opening angle, D1, D2, D3, D4, D5), and its corresponding label = (door opening force).

[0089] Select a suitable neural network model as the first initial network model, for example, select a long short-term memory network LSTM and then train it using a training data set. The specific training steps may be: (1) Input one training data in the training data set into the first initial network model.

[0090] (2) Calculate the door opening force prediction value output by the first initial network model, and calculate the network error (such as mean square error) between the door opening force prediction value and the label corresponding to the training data.

[0091] (3) Use the back propagation algorithm to update the parameters of the first initial network model to optimize the model performance. The goal is to reduce the network error between the output of the first initial network model and the label.

[0092] (4) Repeat the above process until the model converges and obtains the trained door damage detection model.

[0093] The method for constructing a door body damage detection model provided by the present invention benchmarks the simulation model by designing a model test, and at the same time establishes a 3D parameterized damage behavior simulation model that takes damage into account. Finally, the discrete damage behavior simulation model is reduced in order through deep learning to obtain a network model that can realize fast and accurate real-time monitoring of door body damage, thus filling the gap in the industry.

[0094] Based on the content of the above embodiment, as an optional embodiment, the door body core assembly includes a complete door body system and all key subsystems, and the key subsystem includes one or more of a hinge subsystem, a pull rope spring subsystem, a lock subsystem and a sealing subsystem; the hinge subsystem is composed of a skeleton subsystem and an outer door subsystem; Correspondingly, the non-destructive simulation model includes a complete door body system simulation model, a hinge subsystem simulation model, a pull rope spring subsystem simulation model, a lock buckle subsystem simulation model and a sealing subsystem simulation model.

[0095] Taking the door system of a dishwasher as an example, its key subsystems mainly include a hinge subsystem, a pull cord spring subsystem, a lock subsystem, and a sealing subsystem. The above-mentioned key subsystems as a whole are composed of 6 core components: spring (in order to distinguish, the spring of the pull cord spring subsystem is called an expansion spring, and the spring of the lock subsystem is called a locking spring), rope, pulley, hinge wall, door body, door lock buckle and sealing strip.

[0096] Correspondingly, the door system simulation model of the entire dishwasher is formed by integrating the simulation models of five subsystems related to the core components of the door body, namely, the complete door system simulation model, the hinge subsystem simulation model, the pull cord spring subsystem simulation model, the lock subsystem simulation model and the sealing subsystem simulation model.

[0097] The traditional benchmarking method requires obtaining the key mechanical parameters of each door core component through standard mechanical testing. Standard mechanical testing generally requires each door core component to be made into a standard test piece, and it may take multiple tests to complete the acquisition of key mechanical parameters. Many universities or research institutes are unable to complete all the tests, and the test cycle is generally long and the test cost is relatively high.

[0098] In view of this, the present invention benchmarks the door system simulation model of the dishwasher by designing key model tests, and directly performs benchmarking at the system level.

[0099] The following is an introduction on how to build a real benchmarking model for the core components of each door body.

[0100] The hinge subsystem mainly includes the frame and the outer door. In order to be consistent with the non-destructive simulation model constructed subsequently, the state of the benchmark model test is designed as follows: remove the lock, sealing strip and rope, and retain the door hinge.

[0101] The rope-spring subsystem mainly includes the frame, outer door, rope, spring, etc. In order to be consistent with the non-destructive simulation model constructed subsequently, the state of the benchmark model test is designed as follows: remove the lock and sealing strip, and retain the rope, outer door, hinge, etc.

[0102] The lock subsystem mainly includes the frame, outer door and lock. In order to be consistent with the non-destructive simulation model constructed subsequently, the state of the benchmark model test is designed as follows: removing the sealing strip, rope, etc., and retaining the lock, outer door, hinge, etc.

[0103] The sealing subsystem mainly includes the frame, outer door and sealing strip. In order to be consistent with the non-destructive simulation model constructed subsequently, the state of the benchmark model test is designed as follows: remove the lock and rope, and retain the outer door, hinges and sealing strip.

[0104] Taking the door system of the dishwasher as an example, the damage behavior simulation model of each discrete time point is constructed by using the physical benchmarking data and the damage coefficient of each door core component at each discrete time point in the preset time sequence, including but not limited to: Using the physical benchmarking data to benchmark each non-destructive simulation model one by one, to obtain a normal behavior simulation model of each of the door body core components; The normal behavior simulation models are assembled to form a door system simulation model; According to the damage coefficient of each of the door body core components at each discrete time point on the preset time sequence, the door body system simulation model is updated to obtain the damage behavior simulation model under different damage states at each discrete time point.

[0105] The non-destructive simulation model is established based on the theoretical and empirical data of the core components of the door body. It is used to simulate the behavior of the core components of the door body in an undamaged state. The general construction process includes: 1) According to the physical characteristics and mechanical behavior of each door core component, a non-destructive simulation model is constructed. These non-destructive simulation models mainly include: hinge subsystem, pull rope spring subsystem, lock subsystem, sealing subsystem, etc.

[0106] 2) Initialize the model parameters related to each door core component in these non-destructive simulation models, such as the stiffness, strength, friction coefficient, etc. of each component.

[0107] Furthermore, the above-mentioned lossless simulation models are benchmarked using physical benchmarking data to adjust the model parameters of each lossless simulation model so that the simulation output of the lossless simulation model is consistent with the actual test data, and the normal behavior simulation model of each door body core component is obtained.

[0108] As an optional embodiment, in order to simplify and standardize the benchmarking process, when the above-mentioned non-destructive simulation model is benchmarked using physical benchmarking data, the following settings may be made for the corresponding benchmarking process: (1) During the benchmarking test of the non-destructive simulation model of each core component of the door body, since the door body undergoes the same timing action, from the locked state to opening to the preset angle and then returning to the original locked state, relevant data is collected, including recording the magnitude of each door opening force and the corresponding door opening angle at discrete time points.

[0109] (2) The test process uses angle control, that is, the time used to open the door 1° is the same for each door opening and closing process in each test. Through the above tests, the force timing curve and angle timing curve of the real physical prototype can be obtained, which can be used as physical benchmarking data for subsequent simulation benchmarking.

[0110] (3) A force sensor is used to collect the door opening force, and an angle sensor is used to collect the door opening angle. The entire door opening process can be automatically controlled by a robotic arm.

[0111] Furthermore, the normal behavior simulation models of the core components of the door body are combined to form a complete door body system simulation model, specifically including the fusion of the normal behavior simulation models of the core components of the door body, such as the pull rope spring subsystem, the sealing strip subsystem, the lock buckle subsystem, and the hinge subsystem, to form a complete door body system simulation model. The fused door body system simulation model can also be verified at the system level to ensure that the interaction and overall behavior between the core components of the door body conform to the actual physical system.

[0112] The damage coefficient is used to quantify the degree of damage of each door core component at different stages of use. The damage coefficient D of each door core component can be defined based on the fatigue characteristics of the door core component. i The value range of each damage coefficient is [0,1], where 0 indicates no damage and 1 indicates complete failure.

[0113] Then, according to the current opening and closing times of each door core component and the fatigue opening and closing times limit, the damage coefficient at each discrete time point is calculated.

[0114] Furthermore, the relevant model parameters in the door body system simulation model can be updated according to the damage coefficient of each door body core component at each discrete time point. For example, if the damage coefficient of the rope spring subsystem is calculated to be D2 at a discrete time point, the stiffness of the rope spring of the rope spring subsystem in the door body system simulation model can be updated according to the damage coefficient D2.

[0115] In this way, by continuously running within a preset time sequence and obtaining an updated door system simulation model, a damage behavior simulation model at each discrete time point is obtained.

[0116] The method for constructing a door body damage detection model provided by the present invention uses physical benchmarking data to benchmark the non-destructive simulation models of each door body core component, and obtains the normal behavior simulation model of each component. Subsequently, these normal behavior simulation models are combined to form a complete door body system simulation model. Finally, according to the damage coefficient at each discrete time point, the door body system simulation model is updated to obtain the damage behavior simulation model under different damage states. This process solves the technical problem that the simulation software cannot directly realize the establishment of damage evolution, provides a solid foundation for the subsequent construction of the door body damage detection model, and ensures the accuracy and reliability of the model.

[0117] As an optional embodiment, the physical benchmarking data includes force timing curve samples obtained by performing physical opening and closing experiments under angle control on each of the door body core components.

[0118] The following is a detailed description of how to use the physical benchmarking data to benchmark each lossless simulation model one by one to obtain a normal behavior simulation model of each core component of the door body, specifically including: For the lossless simulation model of any core component of the door body, the model parameters of the lossless simulation model are adjusted according to the curve difference between the virtual force timing curve output by the lossless simulation model and the force timing curve sample until the curve difference is within a preset range, and the adjusted lossless simulation model is obtained as the normal behavior simulation model.

[0119] The physical benchmarking data is obtained by conducting physical opening and closing experiments on the core components of the real physical door. Taking the dishwasher door system as the object, the actual collection process can be achieved by the following steps: First, select a representative dishwasher door system and ensure that all its important components (such as pull rope springs, sealing strips, locks, etc.) are in their initial undamaged state.

[0120] In the test, an angle control method was adopted, that is, the angle increment of each door opening remained consistent (for example, 1° each time), and the door opening force data at different opening and closing angles were recorded.

[0121] At the same time, high-precision force sensors and angle sensors are used to collect the timing data of door opening force and door opening angle respectively to form force timing curve samples. These physical benchmarking data will serve as the basis for subsequent simulation model benchmarking.

[0122] The non-destructive simulation model is pre-built using simulation software (such as MotionView) based on the physical properties and mechanical behaviors of each core component of the door body. For the dishwasher door system, it mainly includes the hinge subsystem simulation model, the pull cord spring subsystem simulation model, the lock subsystem simulation model and the sealing subsystem simulation model. The hinge subsystem simulation model is mainly composed of the skeleton subsystem simulation model and the outer door subsystem simulation model.

[0123] The simulation model of the skeleton subsystem is mainly composed of the inner liner, inner liner hinge, door hinge and other components. MotionView and MotionSolve are used as pre- and post-processors respectively. Units are selected according to the force characteristics of each component. All components of the inner liner, inner liner hinge and door hinge are simulated by rigid body units. All components of the inner liner are fixedly connected, the inner liner and inner liner hinge are fixedly connected, the inner liner hinge and the ground are fixedly connected, and the door hinge and the inner liner hinge are set with rotation pairs.

[0124] The simulation model of the outer door subsystem is mainly composed of the inner door, outer door and other components. MotionView and MotionSolve are used as pre- and post-processors respectively. Units are selected according to the force characteristics of each component. The inner door, outer door and other components are simulated with rigid body units. The inner door, outer door and other components are fixedly connected, and the inner door and door hinges are fixedly connected.

[0125] The simulation model of the pull rope spring subsystem is mainly composed of a pull rope, a fixed pulley, a spring group and other components. The unit selection is carried out according to the force characteristics of each component. The fixed pulley is simulated by a rigid body unit, the pull rope is simulated by a nonlinear finite element unit, and the spring is simulated by a spring unit. One end of the pull rope is connected to the door hinge, and the other end of the pull rope is connected to the spring. The spring is fixedly connected to the ground, and contact is defined between the pull rope and the fixed pulley.

[0126] The simulation model of the lock buckle subsystem is mainly composed of the lock buckle, the inner liner lock buckle and other components. The inner liner lock buckle is composed of a lock box, a tension spring and two buckle teeth. MotionView and MotionSolve are used as pre- and post-processors respectively. Units are selected according to the force characteristics of each component. The door lock buckle and the inner liner lock buckle are simulated by rigid body units. The door lock buckle is fixedly connected to the inner door, the inner liner lock buckle is fixedly connected to the inner liner, and the two buckle teeth are connected to the two ends of the tension spring respectively. Each buckle tooth is defined around a revolute pair, and contacts are defined between the buckle teeth, the lock box and the door lock buckle.

[0127] The simulation model of the sealing subsystem can be composed of multiple forces (for example, 34). The macroscopic modeling method is used to divide the entire sealing strip into 34 equal parts. The mechanical behavior of each sealing strip is described by a macroscopic constitutive model. Based on previous test experience, a two-segment linear segment mathematical expression is used to describe the sealing strip macroscopic constitutive model. MotionView and MotionSolve are used as pre- and post-processors, respectively. Units are selected according to the force characteristics of each component. The force of the sealing strip is a pair of action and reaction forces. 34 forces are used, and each force acts on the inner liner and the inner door.

[0128] Furthermore, the physical benchmarking data of each door body core component is used to benchmark the non-destructive simulation model of each door body core component one by one, and the normal behavior simulation model of each door body core component is obtained. Taking the non-destructive simulation model of any door body core component as an example, the following steps can be used to implement it: First, the non-destructive simulation model is run to generate a virtual force time series curve, which describes the change of the door opening force predicted by the non-destructive simulation model at different opening and closing angles.

[0129] Specifically, the Transient solver in MotionSolve can be used for the solution. The outer door subsystem is connected to the skeleton subsystem through a hinge. The revolute pair and the position of the outer door opening are set, and the door body rotation motion is set at the same time. The total analysis time is 20 seconds.

[0130] During the simulation of the non-destructive simulation model, the door body undergoes the same timing action, from the locked state of the door body, opening to the preset angle, and then returning to the original locked state. The simulation adopts angle control, that is, the time used to open the door 1° is the same for each opening and closing process of the non-destructive simulation model. Through the above simulation, the virtual force timing curve and virtual angle timing curve of the virtual non-destructive simulation model can be obtained as the simulation output data. Since the angle control method is adopted, the angle is consistent in the preset timing, so only the virtual force timing curve needs to be paid attention to in the subsequent analysis.

[0131] Then, the virtual force timing curve generated by the non-destructive simulation model is compared with the force timing curve sample obtained from the physical opening and closing experiment of the core component of the door body, and the curve difference between the two is calculated.

[0132] Optionally, the curve difference may be quantified by mean square error (MSE) or other statistical indicators, which will not be described in detail here.

[0133] Furthermore, the model parameters of the lossless simulation model can be adjusted according to the calculated curve difference. The operation can be iterated multiple times until the curve difference is within a preset range.

[0134] For example, if the mean square error between the virtual force timing curve and the force timing curve sample (i.e., the test curve) is less than a certain threshold (such as 0.01N²), the model parameter adjustment is considered to be completed. At this time, the virtual force timing curve output by the non-destructive simulation model is basically consistent with the force timing curve sample of the real physical prototype, that is, the model benchmarking is completed.

[0135] The adjusted non-destructive simulation model is a normal behavior simulation model, which can accurately simulate the behavior of the core components of the door body in an undamaged state.

[0136] By adopting the above method, the lossless simulation model of each door body core component is benchmarked respectively, and the normal behavior simulation model corresponding to each door body core component can be obtained.

[0137] Finally, the present invention can obtain a complete door body system simulation model by integrating the normal behavior simulation models of all door body core components obtained by benchmarking.

[0138] Taking the above-mentioned dishwasher door system as an example, by benchmarking the external door subsystem simulation model, the hinge subsystem simulation model, the pull cord spring subsystem simulation model, the lock buckle subsystem simulation model and the sealing subsystem simulation model, the corresponding normal behavior simulation model of the complete door system, hinge subsystem, pull cord spring subsystem, lock buckle subsystem and sealing subsystem can be obtained. Finally, all these normal behavior simulation models are integrated in the simulation software to obtain the door system simulation model corresponding to the dishwasher door system.

[0139] The method for constructing a door body damage detection model provided by the present invention utilizes physical benchmarking data to benchmark the non-destructive simulation models of each door body core component, and adjusts the model parameters to make the simulation output consistent with the experimental data. Finally, a normal behavior simulation model that can accurately reflect the actual door body behavior is obtained. This process lays a solid foundation for the subsequent construction of a door body damage detection model and effectively improves the reliability of the model.

[0140] Based on the content of the above embodiment, as an optional embodiment, the door body system simulation model is updated according to the damage coefficient of each door body core component at each discrete time point on the preset time sequence, and the damage behavior simulation model under different damage states at each discrete time point is obtained, including: Determine the damage coefficient of any door body core component at any discrete time point based on the current opening and closing times of any door body core component at any discrete time point and the fatigue opening and closing times limit of any door body core component; Determine a scaling factor of a target parameter of any door body core component related to door body damage based on the damage coefficient; At any discrete time point, the door system simulation model is updated according to the scaling coefficients of target parameters of all door core components related to door damage to obtain the damage behavior simulation model at any discrete time point; All the discrete time points are traversed to obtain all the damage behavior simulation models.

[0141] Through the records of the above embodiments, a normal behavior simulation model of the door system is established, which is idealized to be intact throughout the entire time sequence. However, in order to build a high-fidelity digital twin model, it is necessary to consider the damage evolution process of the real physical prototype. How to consider the changes in the structural health status of the door system is a technical difficulty in the industry.

[0142] The present invention proposes reasonable basic assumptions through fatigue damage mechanics theory and establishes a theoretical relationship between the number of door openings and the evolution of structural damage. The following describes in detail how to update the door system simulation model according to the damage coefficient of each door core component at each discrete time point on a preset time sequence to obtain a damage behavior simulation model under different damage states.

[0143] The door system undergoes a sequence of actions during its entire life cycle, which is opening the door and then closing the door, and repeating this process until the end of its service life. In view of this, the present invention proposes several reasonable basic mechanical assumptions: 1) Assuming that the door is opened and closed twice at the same angle, the damage to each door core component is the same. The present invention adopts a special experimental design to make the door open and closed at the same angle each time, so the damage is the same each time. Assuming that the fatigue life of a door core component is N times, the damage each time is opened and closed is one-Nth, which is represented by the D value.

[0144] 2) Assuming that the damage of each core component of the door system can be accumulated linearly, when the damage of a core component of the door system accumulates to a certain extent (such as reaching a certain limit value), the service life of this core component of the door system will be reached. According to Miner fatigue damage mechanics, this limit value can be assumed to be 1.

[0145] 3) In order to perform model reduction sampling later, it is necessary to collect curves with obvious differences in the relationship between door opening force and door opening angle. It is assumed that the relationship between door opening force and door opening angle changes significantly after the door is opened and closed m times. Within m times, it is assumed that the change in the relationship between door opening force and door opening angle is not obvious and can be regarded as the same. The m here can be data such as 100 or 200.

[0146] Through the above assumptions, the relationship between the number of door openings and the process of structural damage evolution can be established. The fatigue life of each door core component under the test design is obtained through simulation or test methods, which can perfectly solve the technical problem that the simulation software cannot directly realize the establishment of damage evolution at the level of the dishwasher door system.

[0147] In the process of constructing the damage behavior simulation model, the object is the complete door system simulation model, which can be solved by the transient solver. The outer door subsystem is connected to the skeleton through a hinge, and a revolute pair is set. The position of the outer door opening is set, and the door body rotation movement is set. The total analysis time is 20s, etc. In the simulation process of the damage behavior simulation model under each damage state, the door system undergoes the same timing action, from the door body locked state, opening to the preset angle, and then returning to the original locked state. The simulation also uses angle control synchronously, that is, during each door opening and closing simulation, the time used to open the door 1° is the same. Through the above simulation, the force timing curve of the virtual simulation prototype under each damage state can be obtained (because the angle control method is adopted, the angle timing curve at this time is actually consistent, so it can be ignored), as the output data of the damage behavior simulation model.

[0148] The steps of obtaining the damage behavior simulation model under different damage states at each discrete time point can generally be divided into: Step 1: For each door core component, at each discrete time point, calculate the damage coefficient based on its current opening and closing times and the fatigue opening and closing times limit. Determine the fatigue opening and closing times limit N of each door core component based on experimental data or theoretical analysis. lim , indicating that the door core component may fail due to fatigue when reaching this number. Record the actual opening and closing times N of each door core component during use cuee , the damage coefficient can be calculated according to the formula: D i =N cuee / N lim Among them, D i It represents the damage coefficient of the core component of the i-th door body, and its value range is [0,1], where 0 represents no damage and 1 represents complete failure.

[0149] Step 2: According to the damage coefficient of each door core component, determine the scaling coefficient of the target parameter related to the door damage, including: determining the key parameters related to the damage of each door core component, such as the stiffness of the pull rope spring, the stiffness of the lock spring, the resilience of the sealing strip, etc. According to the damage coefficient D of each door core component at each discrete sampling point i , the scaling factor of the target parameter can be calculated.

[0150] For example, for the cable-spring subsystem, the scaling factor of the cable stiffness can be expressed as: S i =1-D i , which means that as the damage factor increases, the target parameters (such as the stiffness of the cable) will decrease accordingly.

[0151] Step 3: At each discrete time point, the door system simulation model is updated according to the scaling coefficients of the target parameters of all the door core components to obtain the damage behavior simulation model at the discrete time point. The specific steps include: for each door core component, according to the scaling coefficient S of its target parameter, i , update the corresponding target parameters in the door system simulation model. For example, for the rope spring subsystem, update its initial rope stiffness K0 to K updated :K updated =K0 S i Finally, the door system simulation model can be run using the updated target parameters to obtain the damage behavior simulation model at the discrete time point and output the door opening force time series data.

[0152] Step 4, traverse all preset discrete time points, repeat the above steps, and obtain the damage behavior simulation model at each discrete time point. For example, starting from the first discrete time point, traverse all discrete time points step by step, and at each discrete time point, update the door system simulation model according to the current damage coefficient and the scaling factor of the target parameter, and run the simulation. Collect the results of the damage behavior simulation model obtained at each discrete time point to form a complete set of damage behavior simulation models.

[0153] The present invention determines the damage coefficient based on the current opening and closing times and the fatigue opening and closing times limit of each door body core component, and updates the target parameters in the door body system simulation model according to the damage coefficient, thereby obtaining a damage behavior simulation model under different damage states. This process provides detailed damage state data for the subsequent construction of a door body damage detection model, ensuring that the model can accurately reflect the dynamic damage behavior of the door body system at different use stages.

[0154] Based on the contents of the above embodiments, as an optional embodiment, the target parameters of the hinge subsystem include the deformation parameters at the connection point between the hinge and the pull rope, the target parameters of the pull rope spring subsystem include the pull rope stiffness and the telescopic spring stiffness, the target parameters of the buckle subsystem include the locking spring stiffness, and the target parameters of the sealing subsystem include the sealing strip stiffness.

[0155] The following will take the dishwasher door system as an example to describe the damage of its four subsystems using the corresponding damage coefficients. The damage coefficients of the entire door system simulation model include: 1) Regarding the hinge damage coefficient D1 corresponding to the hinge subsystem: According to the stress analysis of the hinge, the damage is described by the deformation at the connection point between the hinge and the rope. Since the hinge is a rigid body in the model and will not deform, the damage is characterized by the change in the position of the connection point, that is, the deformation parameter at the connection point between the hinge and the rope.

[0156] Figure 4 is a schematic diagram of a method for determining the damage coefficient of a hinge subsystem provided by the present invention, such as Figure 4 As shown in the figure, the vertical line represents the original position of the connection point between the hinge and the pull rope, and the inclined line represents the state position of the connection point between the hinge and the pull rope after deformation. The deformation parameter of the connection point can be expressed as: R θ, then the damage coefficient D1=θ1 / θ0, 0≤θ1≤θ0, 0≤D1≤1, R represents the radial distance from the hinge center to the rope connection point, which is used to convert the angle change θ into the actual displacement.

[0157] 2) Regarding the rope damage coefficient D2 corresponding to the rope spring subsystem: Considering that the drawstring in the drawstring spring subsystem is composed of nylon-wrapped steel wire rope, the force is mainly borne by the steel wire rope, which is a pure tension state. Therefore, in the present invention, the change of the drawstring stiffness will be used to characterize a target parameter for judging damage of the drawstring spring subsystem.

[0158] Specifically, the initial stiffness of the rope is defined as K0, and its stiffness after multiple switching at any discrete time point is defined as K1, then the damage coefficient of the rope spring subsystem target parameter at this discrete time point can be defined as D2 = (K0-K1) / K0. Among them, 0≤K1≤K0, 0≤D2≤1.

[0159] 3) Regarding the rope damage coefficient D3 corresponding to the rope spring subsystem: At the same time, the present invention also considers that the change in the stiffness of the telescopic spring in the rope spring subsystem is also an important target parameter reflecting the damage degree of the subsystem, and its corresponding damage formula can be expressed as: D3=(K2-K3) / K2, where 0≤K3≤K2, and 0≤D3≤1. The initial stiffness of the telescopic spring is K2, and its stiffness after multiple switches at any discrete time point is K3.

[0160] 4) Regarding the sealing strip damage coefficient D4 corresponding to the sealing subsystem: The sealing strip only generates a certain amount of resilience when the door is opened, as it returns to its original shape. After the door is separated from the frame subsystem, the resilience of the sealing strip disappears. Therefore, the determination of the damage coefficient of the sealing subsystem is related to the specific damage behavior simulation model, and its most direct target parameter is the change in the stiffness of the sealing strip.

[0161] Figure 5 The following is a schematic diagram of the stiffness curve of the sealing strip provided by the present invention, wherein the horizontal axis represents the compression deformation of the sealing strip when subjected to an external force, usually expressed in length units (such as millimeters or inches), and the vertical axis represents the force applied to the sealing strip, usually expressed in force units (such as Newtons or pounds). Figure 5 As shown in the figure, the sealing strip stiffness curve describes the force response of the sealing strip under different compression deformations, reflecting the stiffness characteristics of the sealing strip. The slope of the stiffness curve (i.e. the ratio of force to compression deformation) reflects the stiffness of the sealing strip. The larger the slope of the stiffness curve, the harder the sealing strip; the smaller the slope of the stiffness curve, the softer the sealing strip.

[0162] Therefore, when performing the scaling coefficient of the target parameter related to the door body damage of the sealing subsystem at any discrete time point, the present invention quantifies the damage degree of the sealing strip and determines the corresponding damage coefficient D4, where D4=0 indicates that the sealing strip is not damaged, and D4=1 indicates that the sealing strip is completely ineffective.

[0163] Furthermore, the scaling factor of the sealing strip stiffness curve can be calculated according to the damage coefficient D4: S4=1-D4. For example, if D4=0.2, the scaling factor of the sealing strip stiffness curve=0.8.

[0164] Then, at each discrete time point, the sealing strip stiffness curve can be updated according to the scaling factor S4: K5 = K4 × S 4, where K4 is the initial stiffness curve slope of the sealing strip, and K5 is the updated stiffness curve slope.

[0165] By running the door system simulation model using the updated stiffness curve, a damage behavior simulation model at that discrete time point can be obtained.

[0166] 5) Regarding the spring damage coefficient D5 corresponding to the lock subsystem: At the initial moment of opening the door, the door opening force is mainly composed of the locking force generated by the lock, which is mainly the tension caused by the deformation of the locking spring. The damage of the entire lock subsystem can be understood as being mainly caused by the locking spring. Furthermore, the change in the locking spring stiffness in the door system simulation model can be defined to associate it with the corresponding damage coefficient.

[0167] Specifically, the initial stiffness of the locking spring can be defined as K6, the stiffness after multiple switches can be defined as K7, and the damage coefficient D5=(K6-K7) / K6, where 0≤K7≤K6, and 0≤D5≤1.

[0168] Based on the above description, the construction of a 3D parameterized damage behavior simulation model considering damage evolution can be completed. A total of 5 target parameters such as D1-D5 associated with each core component of the door body are set. At each discrete time point, the corresponding damage coefficient D1-D5 is calculated according to the number of door openings. Then, according to the definition of the above damage coefficient, the door system simulation model is automatically updated to obtain corresponding damage behavior simulation models.

[0169] The method for constructing a door damage detection model provided by the present invention determines the damage coefficient based on the current opening and closing times and the fatigue opening and closing times limit of each door core component, and updates the target parameters in the door system simulation model according to the damage coefficient, thereby obtaining a damage behavior simulation model under different damage states. This process provides detailed damage state data for the subsequent construction of the door damage detection model, ensuring that the model can accurately reflect the dynamic behavior of the door system at different use stages.

[0170] Based on the content of the above embodiment, as an optional embodiment, the initial network model is trained using any training data in the training data set to obtain the door damage detection model, including: According to the preset time sequence, the training data corresponding to each discrete time point is sequentially input into the initial network model to obtain a door opening force prediction sequence output by the initial network model; According to the network error between the door opening force prediction sequence and the door opening force label sequence, the model parameters of the initial network model are gradually updated until the model converges, thereby obtaining the door body damage detection model; The door opening force label sequence is composed of the door opening forces at all discrete time points on the preset time sequence.

[0171] Since the detection and calculation of door body damage are performed directly according to all damage behavior simulation models obtained by the methods mentioned in the above embodiments, there are limitations such as low computational efficiency, limited generalization ability and complex data processing. In particular, these door body damage detection models are usually based on physical equations and complex mechanical calculations. Each time the simulation model is run, a lot of computing resources and time are required, and the amount of data generated is usually large. Especially when considering multiple discrete time points and multiple damage states, processing and analyzing these data requires complex algorithms and a large amount of computing resources, which cannot become a digital twin model suitable for actual use.

[0172] The present invention creatively adopts a deep learning network structure algorithm to train the data generated by the 3D parameterized damage behavior simulation model, and outputs a reduced-order version of the damage behavior simulation model that takes damage behavior into account, as a digital twin model of the door system that takes damage evolution into account.

[0173] As an optional embodiment, the present invention selects the Long Short-Term Memory (LSTM) network as the basic architecture of the initial network model, because LSTM can effectively process time series data and capture the time dependency in the input data. However, in actual application, it is also possible to consider using other network models as the basic model architecture for training to obtain a door damage detection model, such as a recurrent neural network (RNN), a gated recurrent unit (GRU), a Transformer network model, or a hybrid model composed of multiple network models as the basic model architecture in scenarios where higher detection accuracy is required.

[0174] Figure 6 This is a schematic diagram of the training process of the door damage detection model provided by the present invention. Figure 6 As shown in the figure, taking LSTM as the basic model architecture as an example, how to implement the training of the door damage detection model is described in detail. It mainly includes but is not limited to the following steps: Step 1: Based on the method provided in the above embodiment, the damage behavior simulation model corresponding to each discrete time point is obtained to generate a training data set using these damage behavior simulation models, and each data sample in the training data set includes input data and label data. The input data is the door opening angle data and the damage coefficient of each door body core component at a discrete time point, and the corresponding label data is the corresponding door opening force data.

[0175] Specifically, taking the door system of a general dishwasher as an example, its entire life cycle will run about 50,000 times. Every 1,000 times (different sampling frequencies are selected according to actual conditions), a door system simulation model is taken. According to the assumptions and methods provided in the above embodiments, the damage value coefficients of each subsystem (core components of the door) of the door system simulation model can be obtained each time. According to the damage coefficient, a damage behavior simulation model corresponding to a door system simulation model can be obtained, and 50 door system simulation models can be obtained. All door system simulation models are input into MotionSolve for solution to obtain the door opening force time series data set of the dishwasher. According to these door opening force time series data sets, a training data set for the improved LSTM deep learning neural network algorithm can be constructed.

[0176] Step 2: Select an LSTM network as the basic architecture of the initial network model. Its input layer can receive the door opening angle and the damage coefficient of each core component of the door body. The LSTM layer processes the time series data and extracts the time-dependent features. Its output layer outputs the predicted door opening force.

[0177] Step 3: Input the training data corresponding to each discrete time point into the initial network model in sequence according to the preset time sequence, including recording the input data and label data of each discrete time point from the training data set, and inputting the input data (door opening angle and damage coefficient) into the initial network model.

[0178] Run the initial network model to obtain the door opening force prediction sequence output by the model, that is, use the initial network model to infer the input data to generate a door opening force prediction sequence (collect the prediction results of each discrete time point to form a complete door opening force prediction sequence).

[0179] Step 4: Calculate the network error based on the difference between the predicted sequence and the label sequence. Obtain the actual door opening force data at each discrete time point from the training data set to form a door opening force label sequence. Use the mean square error (MSE) or other statistical indicators to calculate the network error between the predicted sequence and the label sequence, for example: ; Where N is the total number of discrete time points, and They respectively represent the door opening force error at the i-th discrete time point.

[0180] In step 5, the parameters of the initial network model can be gradually updated according to the calculated network error until the model converges. The specific steps include: 1) Use the back-propagation algorithm to calculate the gradient of the error with respect to the model parameters.

[0181] 2) Use an optimization algorithm (such as Adam) to update the model parameters according to the gradient.

[0182] 3) Repeat the above process until the network error no longer decreases significantly and the model converges.

[0183] Step 6: After the above training process, the parameters of the initial network model are gradually optimized, and finally a high-precision door damage detection model is obtained. The door damage detection model can accurately predict the corresponding door opening force according to the input door opening angle and damage coefficient, thereby realizing real-time detection of the door damage status.

[0184] The method for constructing a door body damage detection model provided by the present invention utilizes a training data set to train the initial network model, gradually optimizes the model parameters, and finally obtains a high-precision door body damage detection model. It can significantly improve computing efficiency, enhance generalization ability, simplify data processing, and meet the needs of real-time monitoring and adaptation to complex working conditions. It is a key step in achieving efficient and accurate door body damage detection.

[0185] Based on the content of the above embodiment, as an optional embodiment, the first initial network model is trained using any training data in the training data set to obtain the door damage detection model, including: According to the preset time sequence, the training data corresponding to each discrete time point is sequentially input into the first initial network model to obtain a door opening force prediction sequence output by the first initial network model; According to the network error between the door opening force prediction sequence and the door opening force label sequence, the model parameters of the first initial network model are gradually updated until the model converges, thereby obtaining the door body damage detection model; The door opening force label sequence is composed of the door opening forces at all discrete time points on the preset time sequence.

[0186] Specifically, the training data set is generated based on the damage behavior simulation model, which takes into account the behavior of the core components of the door body (such as the hinge subsystem, the rope spring subsystem, the lock subsystem, and the sealing subsystem) under different damage states. The damage behavior simulation model is run within a preset time sequence to generate the door opening angle, damage coefficient, and corresponding door opening force data at each discrete time point.

[0187] The door opening angle at each discrete time point and the damage coefficient of each core component of the door body are used as input data, and the door opening force is used as the label data of the input data (the door opening force label sequence is defined as a sequence composed of the door opening forces at all discrete time points on a preset time series). Then, the training data corresponding to each discrete time point can be obtained.

[0188] The first initial network model can adopt a long short-term memory network (LSTM) architecture, which can effectively process time series data and capture time dependencies. It is suitable for the complex mapping relationship between door opening angle, damage coefficient and door opening force in door damage detection tasks.

[0189] According to the preset time sequence, the training data corresponding to each discrete time point is input into the first initial network model in sequence, and the input data includes the door opening angle and the damage coefficient of each door body core component. The first initial network model outputs a door opening force prediction sequence according to the input data. The door opening force prediction sequence represents the predicted value of the door opening force at each discrete time point by the model.

[0190] The door opening force prediction sequence output by the model is compared with the actual door opening force label sequence to calculate the network error, which can be measured by the mean square error (MSE). Based on the calculated network error, the parameters of the first initial network model are updated through the back propagation algorithm. Commonly used optimization algorithms include Adam or RMSprop, which can automatically adjust the learning rate and accelerate model convergence.

[0191] Repeat the above training process until the network error no longer decreases significantly and the model converges. The convergence criterion can be that the error change for multiple consecutive training cycles (epochs) is less than a preset threshold.

[0192] It should be noted that in the above training process, an early stopping mechanism can be used. When the error on the validation set does not improve in multiple consecutive training cycles, the training is stopped to prevent overfitting. The generalization ability of the trained model can also be evaluated through cross-validation (such as k-fold cross-validation) and the optimal model parameters can be selected.

[0193] After the above training process, the parameters of the first initial network model are gradually optimized, and finally a high-precision door damage detection model is obtained. The model can accurately predict the corresponding door opening force according to the input door opening angle and damage coefficient, thereby realizing real-time detection of the door damage status.

[0194] The door body damage detection method provided by the present invention can efficiently train a high-precision door body damage detection model, and provides a reliable technical means for the health management of the door body system.

[0195] Based on the content of the above embodiment, as an optional embodiment, the method of obtaining a door damage recognition agent through reinforcement learning of a door damage detection model includes: A reinforced training set is constructed using the door body damage detection model; any reinforced training data in the reinforced training set includes a door opening angle and a door opening force at any discrete time point, and the label of any reinforced training data is a damage coefficient of each door body core component at any discrete time point, and the door opening force is obtained by inputting the door opening angle and the damage coefficient of each door body core component into the door body damage detection model; The second initial network model is trained using the enhanced training set to obtain the door damage recognition intelligent agent.

[0196] As an optional embodiment, the present invention provides a method for obtaining a door damage recognition agent through reinforcement learning. The method generates a reinforcement training set using a door damage detection model, and trains a second initial network model based on the training set, and finally obtains a door damage recognition agent capable of identifying the door damage state in real time. The method mainly includes but is not limited to the following steps: First, a reinforcement training set needs to be constructed.

[0197] The door damage detection model is built based on deep learning and digital twin technology, and can simulate the behavior of the door system under different damage states. The door damage detection model outputs the corresponding door opening force by inputting the door opening angle and the damage coefficient of each door core component. This model provides a basis for building a reinforced training set.

[0198] Optionally, a reinforced training set is generated using a door damage detection model, and the specific steps include: 1) Select multiple discrete time points within the preset time sequence, which cover the entire process of the door system from no damage to near the end of its service life.

[0199] 2) For each discrete time point, the corresponding door opening angle and the damage coefficient of each core component of the door body are generated as the input part of the enhanced training data.

[0200] 3) The door opening angle and damage coefficient are input into the door damage detection model, and the model outputs the corresponding door opening force as the label part of the enhanced training data.

[0201] In this way, a reinforced training data can be obtained for each discrete time point, whose input is the door opening angle and the damage coefficient of each door core component, and the corresponding label is the door opening force at the same discrete time point. The reinforced training set consists of multiple reinforced training data, each of which corresponds to a discrete time point.

[0202] The obtained enhanced training data set is used to train the second initial network model, so that it can learn the complex mapping relationship between the door opening angle, damage coefficient and door opening force, including: According to the preset time sequence, each enhanced training data in the enhanced training set is input into the second initial network model in turn. The input data includes the door opening angle and the damage coefficient of each door body core component. The second initial network model outputs the predicted door opening force sequence according to the input data.

[0203] The predicted door opening force output by the second initial network model is compared with the label in the reinforced training data, and the network error is calculated. According to the calculated network error, the model parameters of the second initial network model are updated through the back propagation algorithm. The above process is repeated until the network error no longer decreases significantly, the model converges, and the second initial network model is optimized as a door damage recognition agent.

[0204] As an optional embodiment, the first initial network model and the second initial network model are both obtained based on the long short-term memory network model as the basic model architecture.

[0205] The door damage detection method provided by the present invention can efficiently train a door damage recognition intelligent agent, and provides a reliable technical means for the health management of the door system.

[0206] Figure 7 is a schematic diagram of the structure of the door damage detection device provided by the present invention, such as Figure 7 As shown, mainly including but not limited to: An agent training unit 71 is used to obtain a door damage recognition agent through reinforcement learning of a door damage detection model; An intelligent agent recognition unit 72 is used to use the door body damage recognition intelligent agent to recognize the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; The door body detection unit 73 is used to determine the detection result of the door body to be detected based on the damage coefficient sequence of each door body core component.

[0207] It should be noted that, when the door damage detection device provided by the present invention is specifically operated, it can implement the door damage detection method provided by any of the above embodiments, which will not be described one by one here.

[0208] The door body damage detection device provided by the present invention, after establishing a 3D parameterized damage behavior simulation model taking damage into account, reduces the order of the discrete damage behavior simulation model through deep learning to obtain a door body damage detection model, and then learns the door body damage detection model taking into account the damage evolution behavior based on the reinforcement learning algorithm to obtain a door body damage recognition intelligent agent, which can realize real-time monitoring of the health status of the door system.

[0209] In order to fully illustrate the advantages of the method for constructing the door damage detection model provided by the present invention in actual detection, a detailed description is given below in combination with relevant test data.

[0210] Figure 8 This is one of the schematic diagrams showing the performance of the door damage detection model provided by the present invention in the training data set. Fig. 9 This is the second schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set. Fig.10 This is the third schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set. Fig.11 This is the fourth schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set. Fig.12 This is the fifth schematic diagram of the performance of the door damage detection model provided by the present invention in the training data set. The horizontal axis represents the time step, and each unit corresponds to 0.025 seconds. For example, 400 on the horizontal axis represents 10 seconds; the vertical axis represents the numerical value of the predicted value and the actual value, such as the door opening force, the unit is Newton (N). Among them, one of the two force time series curves represents the predicted value, and the other represents the actual value (the dotted line represents the predicted value, and the solid line represents the actual value).

[0211] Among them, Figure 8-Figure 12During the initial network model training process shown in the figure, the total number of door openings (the data scale in the training data set) were 20900, 27100, 34800, 4200 and 9700 respectively. The comparison between the predicted value and the actual value is mainly based on the horizontal axis, which represents the time step and the vertical axis represents the door opening force. The force time series curve fully demonstrates the convergence of the door damage detection model in the training process, which can effectively prove that the door damage detection model obtained after a limited number of trainings has a higher prediction accuracy.

[0212] Fig.13 This is one of the schematic diagrams showing the performance of the door damage detection model provided by the present invention in the test data set. Fig.14 This is the second schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set. Fig.15 This is the third schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set. Fig.16 This is the fourth schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set. Fig.17 This is the fifth schematic diagram of the performance of the door damage detection model provided by the present invention in the test data set. Among them, the horizontal coordinates of the coordinate system where the force time series curve is located represent the time step, and the vertical coordinates represent the value of the door opening force.

[0213] Among them, Figure 13-Figure 17 In the initial network model training process shown, the total number of door openings (data scale in the test data set) were 10400, 11400, 12200, 12900 and 13600 respectively. The convergence of the door damage detection model in the test process is demonstrated mainly through the comparison between the predicted value and the actual value in the two force time series curves, which can help evaluate the generalization ability and prediction accuracy of the model on unseen data. Through the verification of the test data set, it can be determined that the door damage detection model obtained by the present invention has good generalization and high prediction accuracy.

[0214] Fig.18 This is one of the schematic diagrams of the evolution of the loss function value of the door damage detection model provided by the present invention during the training process. Fig.19 This is the second schematic diagram of the evolution of the loss function value of the door damage detection model provided by the present invention during the training process. The horizontal axis represents the number of iterations, the vertical axis represents the loss function value, and the curve is mainly used to represent the change of the loss function value with the increase of the number of training iterations. If the loss function value drops rapidly and tends to be stable, it means that the model converges quickly. The loss function value eventually tends to a smaller value, indicating that the model training effect is good.

[0215] refer to Figure 18-19As shown in the figure, the loss function value Loss error of the door damage detection model at the 171th step reached 2.53, and the convergence accuracy was very high. So far, a deep learning network model with high fitting accuracy was obtained as the digital twin model of the damage behavior of the dishwasher door system.

[0216] The above experiments fully demonstrate the construction method of the door damage detection model provided by the present invention, and obtain an LSTM deep learning network model with high fitting accuracy, which has the ability to effectively utilize timing characteristics and solve longer time series prediction problems.

[0217] Fig. 20 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig. 20 As shown, the electronic device may include: a processor 2010, a communication interface 2020, a memory 2030 and a communication bus 2040, wherein the processor 2010, the communication interface 2020 and the memory 2030 communicate with each other through the communication bus 2040. The processor 2010 may call the logic instructions in the memory 2030 to execute the door body damage detection method, which includes: obtaining a door body damage recognition agent through reinforcement learning of a door body damage detection model; using the door body damage recognition agent to recognize the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and outputting the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; and determining the detection result of the door body to be tested based on the damage coefficient sequence of each door body core component.

[0218] In addition, the logic instructions in the above-mentioned memory 2030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0219] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the door damage detection method provided by the above-mentioned embodiments, and the method includes: obtaining a door damage recognition intelligent agent through reinforcement learning of a door damage detection model; using the door damage recognition intelligent agent to identify the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and outputting the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; based on the damage coefficient sequence of each door body core component, determining the detection result of the door body to be tested.

[0220] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the door damage detection method provided in the above-mentioned embodiments, the method comprising: obtaining a door damage recognition intelligent agent through reinforcement learning of a door damage detection model; using the door damage recognition intelligent agent to identify the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and outputting the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; based on the damage coefficient sequence of each door body core component, determining the detection result of the door body to be tested.

[0221] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0222] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0223] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A door damage detection method, characterized in that: include: Through reinforcement learning of the door damage detection model, the door damage recognition agent is obtained; The door body damage identification intelligent agent is used to identify the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; Based on the damage coefficient sequence of each door body core component, the detection result of the door body to be tested is determined.

2. The door damage detection method according to claim 1, characterized in that: The door damage detection model is constructed by following the steps below: Based on the physical benchmarking data and the damage coefficient of each door body core component at each discrete time point in the preset time sequence, a damage behavior simulation model of each discrete time point is constructed; the physical benchmarking data is obtained by performing physical opening and closing experiments on each door body core component of a real physical door body; The first initial network model is trained using a training data set to obtain the door body damage detection model; any training data in the training data set includes a door opening angle of the damage behavior simulation model at its corresponding discrete time point and the damage coefficient of each door body core component, and the label of the training data is the door opening force at the discrete time point.

3. The door damage detection method according to claim 2, characterized in that: The damage behavior simulation model of each discrete time point is constructed based on the physical benchmarking data and the damage coefficient of each door body core component at each discrete time point in a preset time sequence, including: Using the physical benchmarking data to benchmark each non-destructive simulation model one by one, to obtain a normal behavior simulation model of each of the door body core components; The normal behavior simulation models are assembled to form a door system simulation model; According to the damage coefficient of each of the door body core components at each discrete time point on the preset time sequence, the door body system simulation model is updated to obtain the damage behavior simulation model under different damage states at each discrete time point.

4. The door damage detection method according to claim 3, characterized in that: The physical benchmarking data includes force time series curve samples obtained by performing physical opening and closing experiments under angle control on each of the door body core components; The method of using the physical benchmarking data to benchmark each lossless simulation model one by one to obtain a normal behavior simulation model of each door body core component includes: For the lossless simulation model of any core component of the door body, the model parameters of the lossless simulation model are adjusted according to the curve difference between the virtual force timing curve output by the lossless simulation model and the force timing curve sample until the curve difference is within a preset range, and the adjusted lossless simulation model is obtained as the normal behavior simulation model.

5. The door damage detection method according to any one of claims 3-4, characterized in that: The door body core assembly includes a complete door body system and all key subsystems, wherein the key subsystem includes one or more of a hinge subsystem, a pull rope spring subsystem, a lock buckle subsystem and a sealing subsystem; the hinge subsystem is composed of a skeleton subsystem and an outer door subsystem; Correspondingly, the non-destructive simulation model includes a complete door body system simulation model, a hinge subsystem simulation model, a pull rope spring subsystem simulation model, a lock buckle subsystem simulation model and a sealing subsystem simulation model.

6. The door damage detection method according to claim 5, characterized in that: The normal behavior simulation model is assembled to form a door system simulation model, including: The normal behavior simulation models of the complete door body system, hinge subsystem, pull cord spring subsystem, lock buckle subsystem and sealing subsystem corresponding to the complete door body system simulation model, the hinge subsystem simulation model, the pull cord spring subsystem simulation model, the lock buckle subsystem simulation model and the sealing subsystem simulation model are integrated to obtain the door body system simulation model.

7. The door damage detection method according to claim 6, characterized in that: The method of updating the door system simulation model according to the damage coefficient of each door core component at each discrete time point on the preset time sequence, and obtaining the damage behavior simulation model under different damage states at each discrete time point, includes: Determine the damage coefficient of any door body core component at any discrete time point based on the current opening and closing times of any door body core component at any discrete time point and the fatigue opening and closing times limit of any door body core component; Determine a scaling factor of a target parameter of any door body core component related to door body damage based on the damage coefficient; At any discrete time point, the door system simulation model is updated according to the scaling coefficients of target parameters of all door core components related to door damage to obtain the damage behavior simulation model at any discrete time point; All the discrete time points are traversed to obtain all the damage behavior simulation models.

8. The door damage detection method according to claim 7, characterized in that: The target parameters of the hinge subsystem include the deformation parameters at the connection point between the hinge and the pull rope, the target parameters of the pull rope spring subsystem include the pull rope stiffness and the telescopic spring stiffness, the target parameters of the buckle subsystem include the locking spring stiffness, and the target parameters of the sealing subsystem include the sealing strip rebound force.

9. The door damage detection method according to claim 2, characterized in that: Using any training data in the training data set to train the first initial network model to obtain the door damage detection model includes: According to the preset time sequence, the training data corresponding to each discrete time point is sequentially input into the first initial network model to obtain a door opening force prediction sequence output by the first initial network model; According to the network error between the door opening force prediction sequence and the door opening force label sequence, the model parameters of the first initial network model are gradually updated until the model converges, thereby obtaining the door body damage detection model; The door opening force label sequence is composed of the door opening forces at all discrete time points on the preset time sequence.

10. The door damage detection method according to claim 1, characterized in that: The door damage detection model is reinforced with learning to obtain a door damage recognition agent, including: A reinforced training set is constructed using the door body damage detection model; any reinforced training data in the reinforced training set includes a door opening angle and a door opening force at any discrete time point, and the label of any reinforced training data is a damage coefficient of each door body core component at any discrete time point, and the door opening force is obtained by inputting the door opening angle and the damage coefficient of each door body core component into the door body damage detection model; The second initial network model is trained using the enhanced training set to obtain the door damage recognition intelligent agent.

11. The door damage detection method according to claim 10, characterized in that: The method of training the second initial network model using the enhanced training set to obtain the door damage recognition agent includes: According to the preset time sequence, the enhanced training data corresponding to each discrete time point is sequentially input into the second initial network model to obtain a damage coefficient prediction sequence output by the second initial network model; According to the network error between the damage coefficient prediction sequence and the damage coefficient label sequence, the model parameters of the second initial network model are gradually updated until the model converges, thereby obtaining the door body damage recognition intelligent agent; The damage coefficient label sequence is composed of the damage coefficients of each door core component at all discrete time points on the preset time sequence.

12. The door damage detection method according to claim 11, characterized in that: The first initial network model and the second initial network model are both obtained based on the long short-term memory network model as the basic model architecture.

13. A door damage detection device, characterized in that: include: An agent training unit, used for obtaining a door damage recognition agent through reinforcement learning of a door damage detection model; An intelligent agent recognition unit is used to use the door body damage recognition intelligent agent to recognize the door opening angle sequence and the door opening force sequence of the door body to be tested within a preset time sequence, and output the damage coefficient sequence of each door body core component of the door body to be tested within the preset time sequence; The door body detection unit is used to determine the detection result of the door body to be detected based on the damage coefficient sequence of each door body core component.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the door damage detection method as described in any one of claims 1 to 12 is implemented.

15. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the door damage detection method as described in any one of claims 1 to 12 is implemented.

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the door damage detection method as described in any one of claims 1 to 12 is implemented.

Images