Asynchronous federated learning system and method for turnout switch machine fault diagnosis model

Abstract

The invention provides an asynchronous federal learning system and method for a turnout switch machine fault diagnosis model. The system comprises the following steps: acquiring mechanical vibration signals at three positions of a speed reducer shell, a contact seat and an actuating rod of a switch machine by using a mechanical vibration sensor; information fusion of the vibration signals of the three point locations is carried out; a lightweight dual-scale neural network model; an asynchronous federated learning process resistant to time delay and packet loss; and an asynchronous federated learning system structure. An asynchronous federal learning process resistant to time delay and packet loss is adopted, communication delay and the remaining receiving time of single uploading of an edge computing unit are fully utilized, and the extra computing time for fusion after all units are uploaded is shortened; the asynchronous federated learning system structure clarifies the hierarchical relationship and specification of federated learning of point switch fault diagnosis in a railway system. Distributed training of the switch machine fault diagnosis model is realized, and reliable guarantee is provided for railway transportation safety.

Description

Technical Field

[0001] The present invention relates to the fields of electromechanical equipment fault diagnosis and deep learning distributed training technology, and in particular to railway turnout machine equipment fault diagnosis based on federated learning. Background Art

[0002] The turnout machine is responsible for executing the turnout switching process and determining the running path of the train. However, operating in an open-air environment makes it subject to different weather patterns, temperature fluctuations and humidity levels, and various faults often occur. Algorithms based on signal processing and machine learning rely on prior knowledge and expert experience, which are inefficient, limited in timeliness and high in cost compared to deep learning methods widely used in other fields. The fault diagnosis method of the turnout machine based on data-driven and deep learning shows high accuracy and efficiency. However, the current model distributed training method mainly relies on centralizing data from multiple turnout machines on cloud servers and training fault diagnosis models by using existing data-driven methods. Due to data security and commercial competition, traditional centralized learning methods have limitations in actual operation. Federated learning, as a model training strategy, can achieve model training that keeps data localized rather than centralized aggregation in distributed devices or systems. However, in the currently widely used synchronous federated learning process, when data delays and packet loss occur, the server must spend a considerable amount of time waiting for all packets to arrive before performing aggregate calculations, which leads to high computational costs and significant reduction in accuracy.

[0003] According to the above research, there is currently no special deep learning model designed for switch machine fault diagnosis and a corresponding robust and efficient federated learning system. Summary of the invention

[0004] The embodiments of the present invention provide an asynchronous federated learning system and method for a turnout machine fault diagnosis model, which are used to solve the problems existing in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme.

[0006] The asynchronous federated learning system for the fault diagnosis model of the turnout machine includes: The mechanical vibration sensor is used to: receive the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine in real time; perform real-time information fusion operation on the received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, and high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; The edge computing and storage unit is used to: receive the fused mechanical vibration signal in real time; and train the stored local model in real time based on the received fused mechanical vibration signal; The station-level central server is used to: based on the asynchronous federated learning process with anti-delay and anti-packet loss, receive the trained local model in real time, and perform real-time fusion until the asynchronous federated learning is completed to obtain the railway turnout machine fault diagnosis model; send the railway turnout machine fault diagnosis model to the edge computing and storage unit, so that the edge computing and storage unit uses the railway turnout machine fault diagnosis model to obtain the fault type of the turnout machine; The railway turnout machine fault diagnosis model and local model are lightweight dual-scale neural network models. The lightweight dual-scale neural network model has a large-scale feature extractor and a small-scale feature extractor; the large-scale feature extractor is used to process low-frequency data in mechanical vibration signals, and the small-scale feature extractor is used to process high-frequency data in mechanical vibration signals.

[0007] Preferably, the large-scale feature extractor and the small-scale feature extractor are arranged in parallel; the lightweight dual-scale neural network model also includes a feature stitching layer, a first full convolution module, a second full convolution module, a global average pooling layer, a fully connected layer and an output layer arranged in sequence along the data flow direction; the feature stitching layer is used to perform feature stitching processing on the outputs of the large-scale feature extractor and the small-scale feature extractor.

[0008] Preferably, the large-scale feature extractor and the small-scale feature extractor each have three input channels, corresponding to low-frequency data in the mechanical vibration signal and high-frequency data in the mechanical vibration signal, respectively; Each input channel of the large-scale feature extractor includes a low-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction; each input channel of the small-scale feature extractor includes a high-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction.

[0009] Preferably, the first full convolution module and the second full convolution module respectively include a convolution layer, a batch normalization layer and a ReLU function layer which are sequentially arranged along the data flow direction.

[0010] Preferably, it also includes an advanced server, which is used to: receive the railway turnout and switch machine fault diagnosis model obtained by the station-level central server; fuse the received railway turnout and switch machine fault diagnosis model; update the architecture of the railway turnout and switch machine fault diagnosis model; send the fused / updated railway turnout and switch machine fault diagnosis model to the station-level server, so that the station-level server sends the fused / updated railway turnout and switch machine fault diagnosis model to the edge computing and storage unit.

[0011] Preferably, the edge computing and storage unit is communicatively connected to the station-level server via a UDP communication link, and the station-level server is communicatively connected to the advanced server via a TCP communication link.

[0012] In a second aspect, the present invention provides an asynchronous federated learning method for a turnout machine fault diagnosis model, comprising: Receive the mechanical vibration signals of the reducer housing, contact seat and action rod of the switch machine in real time; The received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine are subjected to real-time information fusion operation to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, as well as high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; Based on the real-time fused mechanical vibration signal, the local model is trained in real time; The local models are integrated in real time until asynchronous federated learning is completed to obtain a railway turnout machine fault diagnosis model; The railway turnout switch machine fault diagnosis model is used for the fault diagnosis of the railway turnout switch machine to obtain the fault type of the railway turnout switch machine.

[0013] It can be seen from the technical solutions provided by the above embodiments of the present invention that the present invention provides an asynchronous federated learning system and method for a turnout machine fault diagnosis model, aiming to solve the problems existing in the existing railway turnout machine fault diagnosis method. The system includes: using a mechanical vibration sensor to collect mechanical vibration signals at three positions of the reducer housing, contact seat, and action rod of the turnout machine; information fusion of vibration signals at three points; a lightweight dual-scale neural network model; an asynchronous federated learning process that is resistant to delay and packet loss; and an asynchronous federated learning system structure. Due to the complex transmission mechanism of the switch machine and the tight coupling of the internal structure, only by placing mechanical vibration sensors at the three points of the reducer housing, the contact seat, and the action rod can the health of the equipment be most effectively reflected; by decomposing the signals at the three points into low-frequency and high-frequency signals and forming channels respectively, the insufficient utilization of information is avoided; the lightweight dual-scale neural network model adopts different feature extraction strategies for low-frequency signals and high-frequency signals respectively, and uses full convolution instead of full connection layer and converts 32-bit floating point weights to 8 bits to achieve lightweight model, reducing the amount of model parameter transmission of federated learning; the asynchronous federated learning process with anti-delay and anti-packet loss makes full use of communication delay and the remaining receiving time of a single upload of the edge computing unit, reducing the additional computing time for waiting for all units to upload and then merge; the asynchronous federated learning system structure clarifies the hierarchical relationship and specifications of the federated learning of fault diagnosis of switch machines in the railway system. The distributed training of the fault diagnosis model of the switch machine is realized, providing reliable guarantee for railway transportation safety.

[0014] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 A logical block diagram of an asynchronous federated learning system for a turnout machine fault diagnosis model provided by the present invention; Figure 2 A schematic diagram of a lightweight dual-scale neural network model of an asynchronous federated learning system for a turnout machine fault diagnosis model provided by the present invention; Figure 3 A schematic diagram of an asynchronous federated learning process of an asynchronous federated learning system for a turnout machine fault diagnosis model provided by the present invention; Figure 4 A schematic diagram of the reduction of additional computing time brought about by asynchronous fusion of an asynchronous federated learning system for a turnout machine fault diagnosis model provided by the present invention; Figure 5 A hierarchical schematic diagram of the structure of the asynchronous federated learning system for the turnout machine fault diagnosis model provided by the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0018] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the field to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0020] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention. Example

[0021] See also Figure 1 The present invention provides an asynchronous federated learning system for a turnout machine fault diagnosis model, comprising: The mechanical vibration sensor 102 is used to: receive the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine in real time; perform real-time information fusion operation on the received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, and high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; The edge computing and storage unit 101 is used to: receive the fused mechanical vibration signal in real time; and train the stored local model in real time based on the received fused mechanical vibration signal; The station-level central server 103 is used to: based on an asynchronous federated learning process that is resistant to delay and packet loss, receive the trained local model (including model parameters) in real time, and perform real-time fusion until the asynchronous federated learning is completed to obtain a railway turnout and switch machine fault diagnosis model; send the railway turnout and switch machine fault diagnosis model to the edge computing and storage unit 101, so that the edge computing and storage unit 101 can use the railway turnout and switch machine fault diagnosis model to perform fault diagnosis to obtain the fault type of the switch machine.

[0022] The railway turnout machine fault diagnosis model and local model are lightweight dual-scale neural network models. The lightweight dual-scale neural network model has a large-scale feature extractor and a small-scale feature extractor; the large-scale feature extractor is used to process the low-frequency data in the mechanical vibration signal (of the reducer housing, contact seat and actuating rod), and the small-scale feature extractor is used to process the high-frequency data in the mechanical vibration signal.

[0023] In the embodiment provided by the present invention, a mechanical vibration sensor is used to collect mechanical vibration signals at three positions of the switch machine, namely, the reducer housing, the contact seat, and the operating rod. This is determined through analysis and research on the operation process of the turnout switch machine, and combined with the actual situation of the high integration of the switch machine, only three points can accurately reflect various abnormal conditions in the operation process of the switch machine. Compared with the three classic status monitoring signals of current, hydraulic pressure, and force, the mechanical vibration of the reducer housing reflects the operating balance of the switch machine motor and the reducer, the vibration of the contact seat reflects the closing of the contact before locking, and the vibration of the operating rod reflects the smoothness and roughness of the switch machine moving the track.

[0024] The information fusion of the vibration signals at the three points mentioned above is to align the vibration signals collected from the reducer housing, contact seat and action rod of the switch machine during one action and pre-process them into low-frequency signals and high-frequency signals, which are used as the three input channels of the low-frequency and high-frequency feature extractors of the neural network model respectively.

[0025] In the embodiment provided by the present invention, the lightweight dual-scale neural network model inputs the fusion features of the above three points and is trained using an end-to-end deep learning model. The low-frequency and high-frequency feature extractors of the model have convolution kernels of different sizes. The large-scale feature extractor is used to process low-frequency data and can capture more macro information, such as trends and global patterns, while the small-scale feature extractor is specifically used to process high-frequency data to capture local characteristics and subtle changes. The lightweight of the model is achieved by using full convolution instead of the fully connected layer and converting the 32-bit floating-point weights to 8 bits.

[0026] In the embodiment provided by the present invention, the asynchronous federated learning process that is resistant to delay and packet loss relies on the server to continuously receive the model parameters uploaded by the edge computing units during the asynchronous federation cycle, and fuse them while receiving. This is different from the synchronous federated learning method that fuses the parameters of all edge computing units after receiving them. It can use the time waiting for other edge computing units to upload and the remaining time for the current unit to upload to perform parameter fusion, which significantly reduces the extra computing time.

[0027] In the preferred embodiment provided by the present invention, Figure 2 As shown, the large-scale feature extractor and the small-scale feature extractor are set in parallel; the lightweight dual-scale neural network model also includes a feature stitching layer, a first full convolution module, a second full convolution module, a global average pooling layer, a fully connected layer and an output layer which are sequentially arranged along the data flow direction; the feature stitching layer is used to perform feature stitching processing on the outputs of the large-scale feature extractor and the small-scale feature extractor.

[0028] In some feasible embodiments, the large-scale feature extractor and the small-scale feature extractor respectively have three input channels, corresponding to low-frequency data in the mechanical vibration signal (of the reducer housing, contact seat and actuating rod) and high-frequency data in the mechanical vibration signal (of the reducer housing, contact seat and actuating rod).

[0029] Each input channel of the large-scale feature extractor includes a low-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction; each input channel of the small-scale feature extractor includes a high-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction.

[0030] In other possible embodiments, Figure 2 As shown, the first full convolution module and the second full convolution module have the same composition, which respectively include a convolution layer, a batch normalization layer and a ReLU function layer arranged in sequence along the data flow direction. Data is input from the convolution layer and output from the ReLU function layer.

[0031] The asynchronous federated learning system provided by the present invention is an application structure based on distributed computing terminals; it is composed of a sensor and data transmission unit for each switch machine, an edge computing unit 101 shared by multiple switch machines, a station-level central server 103, a higher-level advanced server 104, and a communication link (such as Figure 1), where the advanced server can be a single-level main server or multiple levels of servers, forming a pyramid-shaped connection relationship. The edge computing unit is the storage of training data and the trainer of deep learning models, and has exclusive access to the mechanical vibration signals uploaded by multiple connected switch machines after each operation. The station-level central server is the controller of the asynchronous federation cycle and the grassroots unit for the fusion of federated learning parameters. It realizes the reception and update of the edge computing unit training model, and retains the latest model parameters for retransmission and upward transmission. Higher-level main servers are set up by each railway bureau according to demand, and are used for higher-level fusion and model architecture update management of the models of each station-level central server. The communication link between the edge computing unit and the station-level central server uses UDP for transmission, and the station-level and higher-level communications use TCP for transmission.

[0032] The present invention also provides an embodiment for exemplarily showing the working process and effect of the system provided by the present invention.

[0033] In this embodiment, an asynchronous federated learning system for distributed training of a railway turnout switch fault diagnosis model is provided, including using a mechanical vibration sensor 102 to collect mechanical vibration signals from three positions of the reducer housing, contact seat, and action rod of the ZD (J) 9 type switch, information fusion of the vibration signals at the three points, a lightweight dual-scale neural network model, an asynchronous federated learning process that resists delay and packet loss, and an asynchronous federated learning system structure. Using the method provided by the present invention, the mechanical vibration signals collected at the three points are divided into low-frequency and high-frequency signals with 10 Hz as the frequency division point, and fault diagnosis is performed using a lightweight dual-scale neural network model. In order to train the model safely, quickly, and distributedly, an overall federated learning system architecture is constructed based on an asynchronous federated learning process that resists delay and packet loss, and based on the hierarchical relationship of the equipment.

[0034] Figure 2 Schematic diagram of the dual-scale fully convolutional neural network model provided for an embodiment of the present invention. The input is two groups of three-channel time series data, the two groups are low-frequency signals and high-frequency signals respectively, the three channels are the collection results of the three points of the reducer housing, the contact seat, and the action rod, and the output is the classification result of the fault. The dual-scale is to perform convolution calculations of 1 second and 0.1 second for low-frequency signals and high-frequency signals with a frequency division point of 10Hz, respectively, which can help the model obtain effective features in the corresponding frequency band.

[0035] Figure 3A schematic diagram of an asynchronous federated learning process provided for an embodiment of the present invention. In each round of federated learning, the server first distributes the trainable parameters to all clients, and the client loads the trainable parameters into the local model after confirming that the parameters are complete. Thereafter, each client performs model training and parameter updates locally, and after completing a round of local model training, each client sends the trainable parameters of the trained model to the server. The server is responsible for continuously receiving and merging model parameters from all clients during the entire asynchronous federated learning cycle. In the asynchronous federated learning process, parameter fusion is performed in the order of receipt without numbering the clients. Fusion starts with the parameters of the first client that arrives and continues until the end of the federation cycle.

[0036] Figure 4 Schematic diagram of the reduction of additional computation time due to asynchrony provided by an embodiment of the present invention. In scenarios with significant latency, asynchronous fusion uses the time waiting for new data to arrive to update the previous data, thereby achieving a continuous aggregation process. In contrast, synchronous federated learning needs to wait for all updates to be received before starting global aggregation. Therefore, asynchronous fusion exhibits greater flexibility and efficiency in situations with high communication latency. Synchronous federated learning requires additional time when the final parameter update arrives. To aggregate packets, of which represents the time required for the final update aggregation. Asynchronous federated learning takes time, of which Indicates the time to wait for the last round of fusion.

[0037] Figure 5 The hierarchical schematic diagram of the asynchronous federated learning system structure of the turnout machine provided for the embodiment of the present invention is composed of a sensor and data sending unit for each turnout machine, an edge computing unit shared by multiple turnout machines, a station-level central server, a higher-level server, and a communication link. The edge computing unit is the storage of training data and the trainer of the deep learning model, and has exclusive access to the mechanical vibration signals uploaded by the multiple connected turnout machines after each operation. The station-level central server is the controller of the asynchronous federation cycle and the basic unit for the fusion of federated learning parameters. It realizes the reception and update of the edge computing unit training model, and retains the latest model parameters for retransmission and upward transmission. Higher-level servers are set up by each railway bureau according to demand, and are used for higher-level fusion and model architecture update management of the models of each station-level central server. The communication link between the edge computing unit and the station-level central server uses UDP for transmission, and the station-level and higher-level communications use TCP for transmission.

[0038] In a second aspect, the present invention further provides an asynchronous federated learning method for the above system, comprising the following steps: Receive the mechanical vibration signals of the reducer housing, contact seat and action rod of the switch machine in real time; The received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine are subjected to real-time information fusion operation to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, as well as high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; Based on the real-time fused mechanical vibration signal, the local model is trained in real time; The local models are integrated in real time until asynchronous federated learning is completed to obtain a railway turnout machine fault diagnosis model; The railway turnout switch machine fault diagnosis model is used for the fault diagnosis of railway turnout switch machines to obtain the fault type of the switch machine.

[0039] In summary, the present invention provides an asynchronous federated learning system and method for a turnout machine fault diagnosis model, aiming to solve the problems existing in the existing railway turnout machine fault diagnosis method. The system includes: using a mechanical vibration sensor to collect mechanical vibration signals at three positions of the turnout machine's reducer housing, contact seat, and action rod; information fusion of vibration signals at three points; a lightweight dual-scale neural network model; an asynchronous federated learning process that is resistant to delay and packet loss; and an asynchronous federated learning system structure. Due to the complex transmission mechanism of the switch machine and the tight coupling of the internal structure, only by placing mechanical vibration sensors at the three points of the reducer housing, the contact seat, and the action rod can the health of the equipment be most effectively reflected; by decomposing the signals at the three points into low-frequency and high-frequency signals and forming channels respectively, the insufficient utilization of information is avoided; the lightweight dual-scale neural network model adopts different feature extraction strategies for low-frequency signals and high-frequency signals respectively, and uses full convolution instead of full connection layer and converts 32-bit floating point weights to 8 bits to achieve lightweight model, reducing the amount of model parameter transmission of federated learning; the asynchronous federated learning process with anti-delay and anti-packet loss makes full use of communication delay and the remaining receiving time of a single upload of the edge computing unit, reducing the additional computing time for waiting for all units to upload and then merge; the asynchronous federated learning system structure clarifies the hierarchical relationship and specifications of the federated learning of fault diagnosis of switch machines in the railway system. The distributed training of the fault diagnosis model of the switch machine is realized, providing reliable guarantee for railway transportation safety.

[0040] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0041] It can be known from the above description of the implementation mode that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0042] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, 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.

[0043] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An asynchronous federated learning system for a turnout machine fault diagnosis model, characterized in that: include: The mechanical vibration sensor is used to: receive the mechanical vibration signals of the reducer housing, contact seat and action rod of the switch machine in real time; The received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine are subjected to real-time information fusion operation to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, as well as high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; The edge computing and storage unit is used to: receive the fused mechanical vibration signal in real time; and train the stored local model in real time based on the received fused mechanical vibration signal; The station-level central server is used to: based on the asynchronous federated learning process with anti-delay and anti-packet loss, receive the trained local model in real time, and perform real-time fusion until the asynchronous federated learning is completed to obtain the railway turnout switch machine fault diagnosis model; send the railway turnout switch machine fault diagnosis model to the edge computing and storage unit, so that the edge computing and storage unit uses the railway turnout switch machine fault diagnosis model to obtain the fault type of the switch machine; The railway turnout machine fault diagnosis model and local model are lightweight dual-scale neural network models, and the lightweight dual-scale neural network model has a large-scale feature extractor and a small-scale feature extractor; the large-scale feature extractor is used to process low-frequency data in mechanical vibration signals, and the small-scale feature extractor is used to process high-frequency data in mechanical vibration signals.

2. The system according to claim 1, characterized in that The large-scale feature extractor and the small-scale feature extractor are arranged in parallel; the lightweight dual-scale neural network model also includes a feature stitching layer, a first full convolution module, a second full convolution module, a global average pooling layer, a fully connected layer and an output layer arranged in sequence along the data flow direction; the feature stitching layer is used to perform feature stitching processing on the outputs of the large-scale feature extractor and the small-scale feature extractor.

3. The system according to claim 2, characterized in that The large-scale feature extractor and the small-scale feature extractor each have three input channels, corresponding to low-frequency data in the mechanical vibration signal and high-frequency data in the mechanical vibration signal respectively; Each of the input channels of the large-scale feature extractor includes a low-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction; each of the input channels of the small-scale feature extractor includes a high-frequency convolution layer, a batch normalization layer, and a maximum pooling layer arranged in sequence along the data flow direction.

4. The system according to claim 2, characterized in that The first full convolution module and the second full convolution module respectively include a convolution layer, a batch normalization layer and a ReLU function layer which are sequentially arranged along the data flow direction.

5. The system according to claim 1, characterized in that It also includes an advanced server, which is used to: receive the railway turnout machine fault diagnosis model obtained by the station-level central server; integrate the received railway turnout machine fault diagnosis model; and update the architecture of the railway turnout machine fault diagnosis model; The fused / updated railway turnout machine fault diagnosis model is sent to the station-level server, so that the station-level server sends the fused / updated railway turnout machine fault diagnosis model to the edge computing and storage unit.

6. The system according to claim 4, characterized in that The edge computing and storage unit is communicatively connected to the station-level server via a UDP communication link, and the station-level server is communicatively connected to the advanced server via a TCP communication link.

7. An asynchronous federated learning method for a turnout machine fault diagnosis model, characterized in that: include: Receive the mechanical vibration signals of the reducer housing, contact seat and action rod of the switch machine in real time; The received mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine are subjected to real-time information fusion operation to obtain low-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine, as well as high-frequency data of the mechanical vibration signals of the reducer housing, the contact seat and the actuating rod of the switch machine; Based on the real-time fused mechanical vibration signal, the local model is trained in real time; The local models are integrated in real time until asynchronous federated learning is completed to obtain a railway turnout machine fault diagnosis model; The railway turnout switch machine fault diagnosis model is used for the fault diagnosis of the railway turnout switch machine to obtain the fault type of the railway turnout switch machine.

Images