Laser data acquisition and management system, method and device based on cloud service

Abstract

The invention relates to the technical field of laser equipment management and industrial Internet of Things, and discloses a laser data acquisition and management system, method and device based on cloud service. The system comprises an equipment connection layer which is respectively connected with a plurality of lasers and dynamically matches a preset protocol library according to equipment fingerprints corresponding to the lasers so as to convert a communication protocol of each laser into a uniform data format; the edge calculation preprocessing module is deployed at the local end of the laser and is used for performing filtering, compression and feature extraction on original acquisition data of the laser and generating a key parameter set; the access service layer is used for carrying out distributed access processing on the key parameter set; the core processing layer receives the key parameter set and the environmental parameters corresponding to the laser, inputs the key parameter set and the environmental parameters to a time sequence prediction model, and outputs potential faults corresponding to the laser; generating maintenance suggestions corresponding to the potential faults according to the knowledge graph; and the application service layer is used for providing a data interaction interface for a preset user terminal.

Description

Technical Field

[0001] The present application relates to the technical field of laser equipment management and industrial Internet of Things technology, and in particular to a laser data acquisition and management system, method and device based on cloud services. Background Art

[0002] As high-precision industrial equipment, multi-parameter controllable lasers are widely used in cutting, welding, medical and other fields. Their stable operation is crucial to production efficiency and safety. Traditional laser management systems have the following technical bottlenecks:

[0003] 1. Insufficient device compatibility: Lasers from different manufacturers use proprietary communication protocols (such as Modbus, TCP / IP custom protocols, etc.). The existing system requires the development of separate adapter modules for each model, resulting in poor scalability and high deployment costs.

[0004] 2. Inefficient data processing: Raw collected data contains a large amount of noise and redundant information. Directly uploading it to the cloud results in heavy network load and high processing latency. Key features are not extracted, making it difficult to support real-time monitoring and fault prediction. This is especially true for ultrafast lasers, such as femtosecond lasers, where data processing speeds must be as high as possible due to the extremely short duration of a femtosecond.

[0005] 3. Limited fault diagnosis capabilities: Traditional methods rely on threshold judgments or simple rule engines, which cannot capture long-term dependencies between parameters (such as the correlation between power fluctuations and temperature changes) and are insufficient for identifying potential faults in the early stages.

[0006] 4. Delayed maintenance support: After a failure occurs, troubleshooting relies on manual experience, lacks intelligent recommendations based on historical cases, remote collaboration only supports simple video calls, and the maintenance steps are not intuitively integrated with the equipment status, resulting in extended downtime.

[0007] Therefore, a system and method are urgently needed to solve at least one of the above problems. Summary of the Invention

[0008] This application provides a cloud-based laser data acquisition and management system, method, and device, aiming to solve problems such as insufficient equipment compatibility, low data processing efficiency, limited fault diagnosis capabilities, and delayed maintenance support in traditional laser management systems.

[0009] In a first aspect, an embodiment of the present application provides a cloud-based laser data acquisition and management system, including:

[0010] The device connection layer is configured to connect to multiple lasers and dynamically match a preset protocol library based on the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format; the device fingerprint includes identification information and instruction set characteristics;

[0011] An edge computing preprocessing module, deployed on the local side of the laser, is used to filter, compress, and extract features from the raw collected data of the laser to generate a set of key parameters including power, temperature, and operating status;

[0012] an access service layer configured to perform distributed access processing on the key parameter set through a load balancer and a data receiving queue;

[0013] The core processing layer receives the key parameter set and environmental parameters corresponding to the laser and inputs them into a time series prediction model built based on a long short-term memory network to output potential faults corresponding to the laser; constructs a knowledge graph including preset fault characteristics and maintenance cases to generate maintenance suggestions corresponding to the potential faults based on the knowledge graph;

[0014] The application service layer is used to provide a data interaction interface to a preset user terminal, and the user terminal includes a smart device that supports augmented reality remote collaboration and is used to receive the maintenance suggestion.

[0015] In some embodiments, the dynamically matching preset protocol library according to the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format includes: parsing the identification information and instruction set characteristics in the device fingerprint; calling the corresponding protocol adapter from the preset protocol library according to the identification information and instruction set characteristics; the protocol adapter is used to convert the communication protocol of each laser into the unified data format.

[0016] In some embodiments, the raw collected data of the laser are filtered, compressed, and feature extracted to generate a key parameter set including power, temperature, and operating status, including: performing mean filtering or median filtering on the noise signal in the raw collected data according to a preset digital filtering algorithm; using a lossy compression algorithm to compress redundant data in the raw collected data to retain valid data points; extracting the power fluctuation peak and temperature change rate of the raw collected data through time domain feature analysis, and associating them with the operating status identifier corresponding to the raw data to generate the key parameter set.

[0017] In some embodiments, the environmental parameters include temperature and humidity; before the input is to the time series prediction model constructed based on the long short-term memory network, it also includes: training and constructing a time series prediction model based on the historical data corresponding to the laser; during the training process of the time series prediction model, a sliding window method is used to segment the time series data corresponding to the historical data; the time series prediction model is used to capture the long-term dependency between the key parameter set and the environmental parameters through the long short-term memory network unit, and the output includes the component life prediction of the laser and the potential fault of abnormal state warning.

[0018] In some embodiments, the distributed access processing of the key parameter set through a load balancer and a data receiving queue includes: the load balancer is configured to distribute access requests to multiple parallel data processing nodes based on a polling strategy or a load weight strategy; the data receiving queue adopts a distributed message queue architecture to buffer and control the flow of the key parameter set to ensure that the key parameter set accesses the core processing layer at preset time intervals.

[0019] In some embodiments, the construction includes a knowledge graph of preset fault characteristics and maintenance cases, including: performing entity recognition and relationship extraction on preset historical fault reports and maintenance manuals through natural language processing technology to extract the correlation between fault characteristics and maintenance cases; the fault characteristics include fault codes and abnormal parameter thresholds, and the maintenance cases include maintenance steps and laser models; using a graph database to store the fault characteristics and maintenance cases, and constructing a directed graph structure of triples containing fault type, cause and solution to form the knowledge graph.

[0020] Exemplarily, generating maintenance suggestions corresponding to the potential fault based on the knowledge graph includes: converting the potential fault output by the time series prediction model into a knowledge graph query statement; matching multiple similar fault cases corresponding to the knowledge graph query statement in the knowledge graph through a graph search algorithm; performing similarity calculation on each of the similar fault cases based on the model and operation history data of the laser to determine a target similar fault case among the multiple similar fault cases; generating maintenance suggestions including fault cause analysis, spare parts replacement suggestions and operation step instructions based on the target similar fault case.

[0021] In some embodiments, a data interaction interface is provided to a preset user terminal, including: the application service layer establishes a real-time communication channel through the WebSocket protocol, which is used to push real-time curves, potential faults and maintenance suggestions corresponding to the key parameter set to the user terminal; the data interaction interface supports augmented reality remote collaboration function, and is configured to superimpose the first-person perspective image corresponding to the laser and the maintenance suggestions.

[0022] In a second aspect, the present application provides a cloud-based laser data acquisition and management method, which is applied to a cloud-based laser data acquisition and management system as provided in any embodiment of the present application, the method comprising:

[0023] Connecting multiple lasers through a device connection layer, obtaining a device fingerprint corresponding to each laser, wherein the device fingerprint includes identification information and instruction set characteristics, and dynamically matching a preset protocol library according to the device fingerprint to convert the communication protocol of each laser into a unified data format;

[0024] The edge computing pre-processing module deployed at the local end of the laser filters, compresses and extracts features from the original collected data of the laser to generate a key parameter set including power, temperature and operating status;

[0025] Through the access service layer, a load balancer and a data receiving queue are used to perform distributed access processing on the key parameter set;

[0026] The core processing layer receives the key parameter set and environmental parameters corresponding to the laser, inputs them into a time series prediction model built based on a long short-term memory network, and outputs potential faults corresponding to the laser; constructs a knowledge graph including preset fault characteristics and maintenance cases, and generates maintenance suggestions corresponding to the potential faults based on the knowledge graph;

[0027] A data interaction interface is provided to a preset user terminal through an application service layer, and the user terminal includes a smart device that supports augmented reality remote collaboration to receive the maintenance suggestion.

[0028] In a third aspect, the present application provides a cloud-based laser data acquisition and management device, which is applied to a cloud-based laser data acquisition and management system as provided in any embodiment of the present application, the device comprising:

[0029] A protocol conversion unit is configured to connect to multiple lasers through a device connection layer, obtain a device fingerprint corresponding to each laser, the device fingerprint including identification information and instruction set characteristics, and dynamically match a preset protocol library based on the device fingerprint to convert the communication protocol of each laser into a unified data format;

[0030] A parameter generation unit is used to filter, compress and extract features from the original collected data of the laser through an edge computing preprocessing module deployed at the local end of the laser to generate a key parameter set including power, temperature and operating status;

[0031] A service access unit, configured to perform distributed access processing on the key parameter set by using a load balancer and a data receiving queue through an access service layer;

[0032] a suggestion generating unit, configured to receive, through the core processing layer, the key parameter set and environmental parameters corresponding to the laser, input them into a time series prediction model constructed based on a long short-term memory network, and output a potential fault corresponding to the laser; construct a knowledge graph including preset fault characteristics and repair cases, and generate a repair suggestion corresponding to the potential fault based on the knowledge graph;

[0033] The interaction completion unit is used to provide a data interaction interface to a preset user terminal through an application service layer, and the user terminal includes a smart device that supports augmented reality remote collaboration to receive the maintenance suggestion.

[0034] This application dynamically matches a pre-configured protocol library with a device fingerprint (including identification information and instruction set features) to achieve communication protocol conversion for multiple brands and models of lasers. The device fingerprint, as a unique identifier and capability description, enables the system to automatically identify the laser type and match the corresponding communication protocol (such as Modbus, TCP / IP, or custom protocols). This converts raw data from heterogeneous devices into a standardized data format, resolving the compatibility issue of the traditional "one device, one adapter" system.

[0035] By deploying edge computing modules locally on the laser, raw data is filtered (to remove noise), compressed (to reduce data volume), and feature extracted (to refine key parameters) in real time. Focusing on core parameters such as power, temperature, and operating status, digital signal processing algorithms (such as sliding average filtering and wavelet noise reduction) and feature engineering methods are used to reduce redundant data transmission and improve data validity and transmission efficiency.

[0036] Distributed access is achieved using load balancers (such as Nginx and F5) and data reception queues (such as Kafka and RabbitMQ). The load balancer dynamically distributes access traffic to avoid single point overloads. The message queue decouples data reception and processing, supporting peak load shaving and valley filling in high-concurrency scenarios, ensuring system stability and scalability.

[0037] Built on a long short-term memory (LSTM) network, the system inputs key parameter sets (such as power and temperature) and environmental parameters (such as temperature, humidity, and voltage) to capture long-term dependencies in time series data and predict potential faults (such as laser aging and overheating risks). It integrates preset fault characteristics (such as abnormal temperature fluctuation patterns and power attenuation curves) with historical repair cases to construct a fault-cause-solution correlation map. Using graph reasoning algorithms, it quickly generates targeted repair recommendations, achieving intelligent and standardized fault diagnosis.

[0038] The application service layer provides a data interaction interface, supporting access to smart devices (such as AR glasses and smartphones), enabling visual presentation of maintenance recommendations and remote collaboration. Maintenance personnel can use AR devices to obtain real-time fault guidance (such as component location and disassembly and assembly procedures), which, combined with remote guidance from cloud-based experts, improves on-site maintenance efficiency.

[0039] By dynamically matching device fingerprints with protocol libraries, the system is compatible with multiple brands and models of lasers, eliminating the need to develop adapter modules for each device type, reducing integration costs and improving system versatility. Edge computing preprocessing reduces raw data volume (by filtering noise and compressing redundancy), alleviating network transmission pressure and cloud computing load. The distributed access architecture supports high-concurrency data access, avoiding performance bottlenecks in traditional centralized systems and improving real-time data processing. The LSTM model uses time series data to mine potential fault characteristics (such as performance degradation trends), enabling earlier detection of anomalies compared to traditional threshold-based alarms. The knowledge graph integrates historical experience to quickly locate fault causes and accurately recommend repair solutions, addressing the limitations of traditional systems that only provide alarms but no diagnosis. The AR remote collaboration function breaks geographical restrictions, enabling real-time interaction between on-site personnel and experts, shortening the troubleshooting cycle. The intelligent generation of repair recommendations reduces manual judgment costs, improves maintenance efficiency and standardization, and reduces downtime losses. A complete closed loop is formed from device access, data processing, fault prediction, and maintenance support. Accumulated historical data can be used to reversely optimize the time series model and knowledge graph, enabling continuous iteration of system capabilities and providing data support for the entire laser lifecycle management. The cloud-based layered architecture (device connection layer, edge layer, core processing layer, and application layer) supports modular expansion, facilitating the subsequent connection of more types of devices or new functions (such as energy consumption analysis and production capacity optimization), adapting to the diverse needs of industrial Internet scenarios.

[0040] In summary, the system achieves equipment compatibility, efficient data processing, intelligent diagnosis and rapid maintenance support through technological integration, specifically solving the core pain points of traditional laser management systems and promoting the upgrade of laser equipment management to intelligence and service.

[0041] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0043] Figure 1This is a schematic structural diagram of a cloud-based laser data acquisition and management system provided by one embodiment of the present application;

[0044] Figure 2 This is a schematic flow chart of the steps of a cloud service-based laser data collection and management method provided in one embodiment of the present application;

[0045] Figure 3 This is a structural diagram of a cloud service-based laser data acquisition and management device provided in one embodiment of the present application.

[0046] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0048] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0049] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items having substantially the same functions and effects. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or order of execution, and that terms such as "first" and "second" do not necessarily define differences.

[0050] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0051] It will also be understood that the term "and / or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0052] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0053] As high-precision industrial equipment, multi-parameter controllable lasers are widely used in cutting, welding, medical and other fields. Their stable operation is crucial to production efficiency and safety. Traditional laser management systems have the following technical bottlenecks:

[0054] 1. Insufficient device compatibility: Lasers from different manufacturers use proprietary communication protocols (such as Modbus, TCP / IP custom protocols, etc.). The existing system requires the development of separate adapter modules for each model, resulting in poor scalability and high deployment costs.

[0055] 2. Inefficient data processing: Raw collected data contains a large amount of noise and redundant information. Directly uploading it to the cloud results in heavy network load and high processing latency. Key features are not extracted, making it difficult to support real-time monitoring and fault prediction. This is especially true for ultrafast lasers, such as femtosecond lasers, where data processing speeds must be as high as possible due to the extremely short duration of a femtosecond.

[0056] 3. Limited fault diagnosis capabilities: Traditional methods rely on threshold judgments or simple rule engines, which cannot capture long-term dependencies between parameters (such as the correlation between power fluctuations and temperature changes) and are insufficient for identifying potential faults in the early stages.

[0057] 4. Delayed maintenance support: After a failure occurs, troubleshooting relies on manual experience, lacks intelligent recommendations based on historical cases, remote collaboration only supports simple video calls, and the maintenance steps are not intuitively integrated with the equipment status, resulting in extended downtime.

[0058] Therefore, a system and method are urgently needed to solve at least one of the above problems.

[0059] To resolve the above, please refer to Figure 1The embodiment of the present application provides a cloud-based laser data acquisition and management system, the system comprising: a device connection layer, which connects to multiple lasers respectively and dynamically matches a preset protocol library based on the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format; the device fingerprint includes identification information and instruction set features; an edge computing preprocessing module, deployed on the local end of the laser, for filtering, compressing and feature extracting the raw collected data of the laser to generate a key parameter set including power, temperature and operating status; an access service layer, configured to perform distributed access processing on the key parameter set through a load balancer and a data reception queue; a core processing layer, which receives the key parameter set and the environmental parameters corresponding to the laser and inputs them into a time series prediction model constructed based on a long short-term memory network to output potential faults corresponding to the laser; a knowledge graph including preset fault features and maintenance cases is constructed to generate maintenance suggestions corresponding to the potential faults based on the knowledge graph; and an application service layer, which provides a data interaction interface to a preset user terminal, the user terminal including a smart device supporting augmented reality remote collaboration, for receiving the maintenance suggestions.

[0060] Specifically, this system integrates device protocol adaptation, edge computing preprocessing, time series prediction models, and knowledge graph technology through a layered architecture design to address the compatibility, data processing efficiency, fault diagnosis, and maintenance support issues of traditional laser management systems. The system architecture includes five core modules, and the technical content of each layer is as follows:

[0061] 1. Device connection layer: A dynamic protocol matching mechanism based on device fingerprints. Device fingerprints contain unique identification information (such as device ID, manufacturer model, hardware version) and instruction set characteristics (such as communication port, data frame format, and instruction interaction logic), and are generated through handshake data collected during device initialization. Pre-built protocol library: Pre-integrated mainstream laser protocols (Modbus, TCP / IP custom protocols, etc.), each with corresponding parsing rules and data mapping templates (such as mapping the "power register address" in a private protocol to a "power_value" field in a unified format). Dynamic matching process: When a device is connected, the protocol library is retrieved through the device fingerprint, and the corresponding parsing rules are automatically loaded to convert the raw data from different manufacturers into a unified JSON format that contains basic device information, real-time parameters, and status codes.

[0062] 2. Edge Computing Preprocessing Module: Real-time data processing is achieved by deploying edge computing nodes (such as industrial-grade PLCs or embedded gateways) locally at the laser. Sensor noise (such as high-frequency fluctuations in temperature sensors) is removed using algorithms such as sliding average filtering and Kalman filtering. Principal component analysis (PCA) or wavelet transforms are used to compress redundant data and extract key parameters (such as power mean / fluctuation amplitude, temperature gradient, operating time, and number of abnormal shutdowns), achieving a data compression ratio of 60%-80%. In the event of a network interruption, preprocessed data is temporarily stored and uploaded in batches upon network restoration to ensure data integrity.

[0063] 3. Access service layer: This layer utilizes a microservices architecture, enabling high-concurrency data access through load balancers (such as Nginx) and message queues (such as Kafka). Data traffic is dynamically distributed based on node load to avoid overloading a single point. Data collection and processing are decoupled, supporting asynchronous processing and buffering data during peak traffic to reduce system pressure. Incoming data is formatted and checked for integrity, filtering out invalid data.

[0064] 4. Core processing layer: Builds a multi-parameter correlation model based on a long short-term memory (LSTM) network. Input data: Key parameter sets (power, temperature, operating status) + environmental parameters (workshop temperature and humidity, power grid fluctuation data), with a time window set to 72 hours to capture long-term dependencies. Model training: Uses historical fault data to annotate samples, optimizes model parameters through backpropagation, and outputs potential fault probabilities (e.g., confidence levels for fault types such as "power module aging" and "cooling system abnormality").

[0065] Knowledge Graph Construction: Entities and Relationships: Entities include fault type, fault characteristics (e.g., "continuous temperature rise + power fluctuation" corresponds to "cooling fan failure"), repair cases (steps, tools, spare parts list), and device models. Relationships are defined as "fault-feature association," "case-fault correspondence," and "spare parts-device matching." Reasoning Mechanism: Based on the predicted fault type, the knowledge graph is searched to generate repair recommendations (e.g., recommended repair steps, spare part models, and historical success rates for similar cases).

[0066] 5. Application service layer: Provides a web-based monitoring interface and mobile API, supporting real-time data visualization (trend curves, device status heat maps), historical data query, and early warning push (SMS / email).

[0067] Augmented Reality (AR) Remote Collaboration: Maintenance personnel connect to the system through smart glasses (such as HoloLens) and receive real-time fault location annotations and repair step instructions (pictures, text, and video) overlaid on the 3D model of the equipment. This allows for real-time annotation guidance from remote experts, and displays real-time equipment parameters (such as temperature changes) during the repair process, ensuring that repair steps are aligned with equipment status.

[0068] In some embodiments, the specific workflow corresponding to the system includes:

[0069] 1. Device Access and Protocol Adaptation Process: Initialization Phase: After the laser is powered on, a handshake signal is sent to the device connection layer, carrying basic device information (manufacturer, model). The system uses probe commands (such as "Query Communication Protocol Version") to collect instruction set features and generate a device fingerprint. The system then matches the pre-configured protocol library. If a corresponding protocol exists, parsing rules are automatically loaded. If the protocol is new, a manual annotation process is triggered (engineers configure parsing rules and expand the protocol library through a visual interface). Data Collection Phase: Raw data is collected at a fixed interval (e.g., 100ms), converted to a unified format by the protocol parsing module, and output to the edge computing node.

[0070] 2. Edge computing preprocessing: Raw data is first filtered through a digital filter to remove high-frequency noise. For example, a ±5°C fluctuation threshold is set for temperature data; data exceeding this threshold triggers secondary filtering. A sliding window (with a window size of 50 sampling points) is used to calculate the power mean and standard deviation, which serve as a "power stability" feature. The compressed data is encrypted and uploaded to the access service layer via the MQTT protocol. In the event of a network interruption, it is stored in a local SQLite database and subsequently transmitted upon recovery.

[0071] 3. Core Processing Layer Model Operation Mechanism: Time Series Forecasting Model: Automatically synchronize the latest seven days of historical data (including normal and faulty samples) every morning, and incrementally train the model to avoid time-consuming repetitive training. Real-time input data is aggregated into 10-minute windows (for example, to calculate the temperature gradient: current temperature minus the temperature 10 minutes ago) and normalized before input into the model.

[0072] Knowledge Graph Update: When a new fault case is added, maintenance reports are parsed using natural language processing (NLP), entities and relationships are extracted, and the graph is updated after manual review. This allows maintenance personnel to annotate new features in real time during AR collaboration and provide feedback to the graph construction module.

[0073] 4. AR Remote Collaboration Implementation Steps: When a fault occurs, the system sends a warning message to maintenance personnel, including the predicted fault type and preliminary steps recommended by the knowledge graph. The maintenance personnel then scan the device's QR code while wearing an AR device. The system then loads a 3D model of the device and highlights the predicted faulty component (e.g., a cooling fan in red). The remote expert then uses the web client to view real-time device parameters and AR-viewed footage, annotating key repair points (e.g., "Remove this screw, replace the filter") and simultaneously recording these steps in the maintenance case library.

[0074] The provided system has the following benefits:

[0075] 1. Resolve device compatibility issues and reduce deployment costs: By dynamically matching device fingerprints to the protocol library, there's no need to develop separate adapter modules for new devices. This reduces the time it takes to connect new devices from 2-3 days to 30 minutes, reducing deployment costs by over 70%. The protocol library supports custom extensions and is compatible with future laser models, significantly improving system scalability.

[0076] 2. Improved data processing efficiency and support for real-time monitoring: Edge computing preprocessing reduces invalid data uploads by over 60%, lowering network bandwidth usage and shortening cloud processing latency from seconds to under 100ms. Key feature extraction focuses on core parameters (such as the correlation between power fluctuations and temperature), providing high-quality input for real-time monitoring and model prediction, and improving decision response speed by 50%.

[0077] 3. Enhanced fault diagnosis capabilities and early warning: The LSTM model captures long-term dependencies between parameters (for example, predicting a cooling failure when the temperature rises by 0.5°C per hour for 12 consecutive hours and the power drops by 1%). Compared to traditional threshold methods, this model improves early fault identification by 40%. The knowledge graph integrates historical cases and supports multi-dimensional reasoning (for example, recommending the most likely fault cause based on equipment age and fault feature combinations), reducing missed diagnosis by 30%.

[0078] 4. Optimizing maintenance support and reducing downtime: AR remote collaboration enables real-time linkage between maintenance steps and equipment status (for example, simultaneous display of temperature change feedback during component replacement), reducing maintenance time by over 40%. Intelligently recommending maintenance cases and spare parts lists reduces reliance on manual experience, improving troubleshooting efficiency for novice maintenance personnel by 60%.

[0079] 5. System-level benefits: The overall solution utilizes a layered, decoupled design, enabling flexible scalability (the access service layer can scale horizontally to accommodate device growth), adapting to the needs of factories of varying sizes. Unified data management lays the foundation for advanced applications such as digital twins and energy optimization, establishing a closed-loop management system for the entire laser lifecycle.

[0080] In summary, the system solves the four major bottlenecks of traditional laser management through the integration of protocol adaptation, edge computing, intelligent models and AR technology, achieving "plug and play" equipment access, "precise and efficient" data processing, "intelligent prediction" fault diagnosis, and "virtual and real integration" maintenance support, significantly improving the reliability and operation and maintenance efficiency of industrial lasers, and has high engineering practicality and industry promotion value.

[0081] In some embodiments, the dynamically matching preset protocol library according to the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format includes: parsing the identification information and instruction set characteristics in the device fingerprint; calling the corresponding protocol adapter from the preset protocol library according to the identification information and instruction set characteristics; the protocol adapter is used to convert the communication protocol of each laser into the unified data format.

[0082] When a laser is connected to the system, the device connection layer first parses the device fingerprint. This identification information includes the device's unique ID, manufacturer name, model, and hardware version number. Instruction set features include the communication port (such as the COM3 port in Modbus RTU), data frame format (such as the start character 0x01 and the location of the data length field), and handshake instructions (such as the hexadecimal encoding of "Query device status" 0x03 0x00 0x01 0x00 0x02 0xCRC"). This information is dynamically obtained from device initialization messages (such as the manufacturer-defined header carried in the TCP three-way handshake) through regular expression matching and field extraction algorithms to generate a standardized device fingerprint data structure (JSON format).

[0083] The pre-built protocol library stores parsing rules for various vendor protocols. Each protocol corresponds to a protocol adapter (such as a Java class or Python module). This adapter includes data parsing functions (converting hexadecimal raw data into decimal parameters), field mapping tables (for example, mapping the "0x0002 register" in a proprietary protocol to the "temperature" field in a standardized format), and error checking logic (such as a CRC checksum algorithm). The system searches the protocol library based on the vendor and model information in the device fingerprint and matches it to the corresponding adapter. If the instruction set characteristics differ from the pre-built protocol (for example, a different port number), the custom port number is dynamically loaded through the adapter's parameter configuration interface, eliminating the need for code modifications.

[0084] The protocol adapter receives raw byte stream data, splits the data frame according to the parsing rules, extracts the payload (such as power value and operating status code), and converts it into a unified format according to the field mapping table (for example, converting power units (W or mW) from different manufacturers to W). The final output is JSON data containing basic device information, real-time parameters, and timestamps (such as {"device_id":"1001","power":200.5,"status":"normal","timestamp":"2023-10-01 12:00:00"}).

[0085] Eliminating the need to develop separate adapter modules for each model, new devices only require configuration in the protocol library (non-coding), reducing deployment costs by over 60% and shortening the adaptation cycle from two weeks to two hours. Automatically identifying protocol types through device fingerprinting supports mixed integration of devices from multiple vendors (such as TRUMPF, IPG, and Han's Laser), improving system scalability. Adapting new laser models requires only updating the protocol library, eliminating the need to restart services.

[0086] In some embodiments, the raw collected data of the laser are filtered, compressed, and feature extracted to generate a key parameter set including power, temperature, and operating status, including: performing mean filtering or median filtering on the noise signal in the raw collected data according to a preset digital filtering algorithm; using a lossy compression algorithm to compress redundant data in the raw collected data to retain valid data points; extracting the power fluctuation peak and temperature change rate of the raw collected data through time domain feature analysis, and associating them with the operating status identifier corresponding to the raw data to generate the key parameter set.

[0087] Preset digital filtering algorithms include mean filtering (window size 5, for example, averaging five consecutive temperature samples to remove high-frequency jitter) and median filtering (suitable for pulse noise, such as occasional abnormal extreme values output by sensors). The system automatically selects the filtering method based on the data type: mean filtering is used for slowly varying parameters such as temperature and humidity, while median filtering is used for switching states (such as "emergency stop" signals).

[0088] During the filtering process, dynamic thresholds are set (such as the normal temperature range [20°C, 30°C]). Abnormal data exceeding the threshold triggers secondary verification (three consecutive samples are taken to confirm whether it is truly abnormal) to avoid incorrect filtering of valid signals.

[0089] A lossy compression algorithm (such as piecewise linear fitting, which thins out the power curve and retains key turning points) is used, with a compression accuracy of 5% (i.e., the maximum error between the compressed and original data does not exceed 5%). For example, if the original power data has 100 samples per second, 20 characteristic points (peaks, valleys, and stable segment averages) will be retained after compression. One-hot encoding is performed on the operating status identifiers (such as "normal operation," "standby," and "fault") and merged with the compressed numerical data to form a structured dataset.

[0090] By calculating the peak power fluctuation (maximum power minus minimum power in the current cycle) and the temperature change rate ((current temperature minus temperature 10 minutes ago) / 600 seconds), the feature values are updated in real time using a sliding window (30-minute window length). The extracted features are timestamped and associated with the operating status indicator (for example, the peak power fluctuation at the corresponding moment in the "fault" state). This generates a key parameter triple consisting of "peak power fluctuation peak value - temperature change rate - operating status," for example, {"peak_power":5.2,"temp_rate":0.3,"status":"alarm"}.

[0091] After noise filtering, data efficiency increased from 70% to 95%, compressed data volume decreased by 70%, network transmission latency was reduced by 40%, and cloud processing pressure was significantly reduced. Time-domain analysis extracts core indicators reflecting device status (such as the correlation between power fluctuations and fault conditions), providing high-value input for subsequent fault prediction models and avoiding the "garbage in, garbage out" problem.

[0092] In some embodiments, the provided system uses a lightweight communication protocol (such as MQTT or a custom UDP protocol) to reduce data packet overhead and improve the transmission efficiency of laser status data (such as power, temperature, and wavelength).

[0093] In some embodiments, the corresponding configuration parameters of the system are shown in the following table:

[0094]

[0095] In some embodiments, the system's corresponding exception handling mechanisms include: communication interruption handling: initiating a reconnection mechanism (initial interval 1 second, exponential backoff up to 60 seconds); local caching of unsent data (up to 100 entries); and data anomaly handling: triggering three levels of alarms (logging → email notification → SMS alert); and automatically isolating abnormal devices.

[0096] In some embodiments, the provided system features a layered security mechanism, including: Role-Based Access Control (RBAC): This divides permissions into multiple levels, such as operator, engineer, and administrator, limiting the scope of execution for critical commands (such as power adjustments). End-to-end encryption: This utilizes SM4 or AES-256 encryption for data transmission, combined with digital certificates to verify device identity and prevent man-in-the-middle attacks. Operational audit logs: This records all remote operation commands and parameter modifications, and supports blockchain-based authentication to ensure that the logs cannot be tampered with.

[0097] In some embodiments, the workflow corresponding to the provided system includes: data acquisition phase: Step 1: After the laser is powered on, it sends an identity authentication message (including device ID and MAC address); Step 2: The server responds to the authentication request and establishes a persistent connection; Step 3: Sending data packets at 100ms intervals (including 12 parameters such as temperature, power, and operating status).

[0098] Data processing phase: Step 1: The raw data enters a circular buffer (set to 1MB); Step 2: The parsing thread extracts the complete data frame from the buffer; Step 3: Passes three-level validation (format / range / logic); Step 4: Converts to a unified JSON format and stores it in Redis.

[0099] In some embodiments, the laser includes a network-type laser and a serial-port laser. The network-type laser is connected to the Gigabit Ethernet port of the corresponding device connection layer of the server through an RJ45 interface; the serial-port laser is connected to the USB3.0 port of the corresponding device connection layer of the server through an RS485 to USB adapter; the host computer establishes a Socket connection with the device connection layer through the TCP / IP protocol (the port number is set to 502 by default); the network interface supports the IEEE 802.3af PoE standard; the RS485 interface supports ±15kV ESD protection; the server is configured with at least 4-core CPU / 16GB memory / dual Gigabit network cards.

[0100] In some embodiments, the environmental parameters include temperature and humidity; before the input is to the time series prediction model constructed based on the long short-term memory network, it also includes: training and constructing a time series prediction model based on the historical data corresponding to the laser; during the training process of the time series prediction model, a sliding window method is used to segment the time series data corresponding to the historical data; the time series prediction model is used to capture the long-term dependency between the key parameter set and the environmental parameters through the long short-term memory network unit, and the output includes the component life prediction of the laser and the potential fault of abnormal state warning.

[0101] Collect historical laser operation data (at least one year, covering normal operation, abnormal warnings, and failure shutdown scenarios), annotating the fault type (e.g., "power module aging" or "cooling fan failure") and the time of occurrence. Use a sliding window approach to segment the time series data (with a window size of 24 hours and a step size of 1 hour). Each window contains a 24-hour set of key parameters (power, temperature, operating status) and environmental parameters (workshop temperature and humidity). This forms the input sample [X1, X2, ..., Xn], labeled with whether a fault will occur within the next hour and the fault type.

[0102] The time series prediction model is built on an LSTM network, consisting of two layers of LSTM units (128 neurons per layer) and one fully connected layer (outputting fault type probabilities). The Adam optimizer is used, and the loss function is cross-entropy (for multi-classification scenarios). Dropout regularization (with a dropout rate of 0.2) is introduced during training to prevent overfitting. The model's F1 score on the test set is validated every 100 epochs, and training is terminated when it reaches or exceeds 0.9. Real-time input data is dynamically scrolled using a consistent window size (24 hours). Missing data is supplemented using linear interpolation, and normalization (Z-score standardization) is performed before input. The LSTM unit uses memory cells to capture long-term dependencies between parameters (for example, if the temperature rises by 1°C per day for three consecutive days and the power decreases by 0.5%, it predicts that the cooling system is about to fail). Outputs include potential failure predictions (e.g., "The remaining life of the pump source is 300 hours") and abnormality warnings (e.g., "Abnormal power fluctuations may occur in the next two hours").

[0103] Compared to traditional threshold methods (which only determine whether a single parameter exceeds a limit), the LSTM model can identify gradual trends in parameter combinations (for example, the weakening of the negative correlation between temperature and power), increasing the early fault identification rate from 50% to 80%. Component life prediction allows users to stock spare parts in advance to avoid unplanned downtime. Abnormal status warnings enable maintenance personnel to intervene before a fault occurs, reducing downtime losses by over 30%.

[0104] In some embodiments, the distributed access processing of the key parameter set through a load balancer and a data receiving queue includes: the load balancer is configured to distribute access requests to multiple parallel data processing nodes based on a polling strategy or a load weight strategy; the data receiving queue adopts a distributed message queue architecture to buffer and control the flow of the key parameter set to ensure that the key parameter set accesses the core processing layer at preset time intervals.

[0105] The load balancer supports two distribution strategies: Round-robin strategy: evenly distributes access requests to multiple data processing nodes in sequence (for example, with three nodes, the first request goes to node 1, the second to node 2, the third to node 3, and the fourth back to node 1). This strategy is suitable for scenarios with consistent node performance. Load weight strategy: dynamically adjusts the weight based on the node's current CPU usage and memory usage (for example, if node A is loaded at 80%, the weight is 0.5, and if node B is loaded at 30%, the weight is 1.5), prioritizing requests to low-load nodes to avoid overload. The load balancer uses a heartbeat detection mechanism (sending a health check packet every 5 seconds) to eliminate faulty nodes and ensure high availability.

[0106] Distributed message queue implementation: The data receiving queue uses a Kafka or RabbitMQ architecture. Each laser's key parameter set is sent as a message to a topic, supporting partitioning (e.g., partitioning by device manufacturer to improve consumption efficiency). Queues are configured with flow control parameters: maximum throughput (e.g., 10MB / s) and message retention time (72 hours by default). When incoming traffic surges, unprocessed messages are temporarily stored in the queue. The core processing layer pulls messages in batches at preset intervals (e.g., 50ms) to avoid instantaneous pressure on the system.

[0107] Supporting simultaneous access of over 2,000 lasers, the system throughput increases by 50% during peak traffic, and node load variations are kept within 10%, preventing service interruptions caused by single-point overload. The message queue buffers bursty data, ensuring stable operation of the core processing layer and preventing data loss during network jitter (persistent queue storage), enhancing the system's fault tolerance.

[0108] In some embodiments, the construction includes a knowledge graph of preset fault characteristics and maintenance cases, including: performing entity recognition and relationship extraction on preset historical fault reports and maintenance manuals through natural language processing technology to extract the correlation between fault characteristics and maintenance cases; the fault characteristics include fault codes and abnormal parameter thresholds, and the maintenance cases include maintenance steps and laser models; using a graph database to store the fault characteristics and maintenance cases, and constructing a directed graph structure of triples containing fault type, cause and solution to form the knowledge graph.

[0109] By using natural language processing (NLP) to process historical fault reports and repair manuals, named entity recognition (NER) was used to extract fault feature entities (such as the fault code "E001" and the abnormal parameter threshold "temperature > 35°C") and repair case entities (such as the repair step "replace filter element" and the applicable laser model "Dazs G4020"). A relational extraction model identified associations between entities (such as "fault code E001 corresponds to the fault type 'cooling fan stopped'" and "repair step 'replace filter element' applies to model G4020"). A rule engine (regular expressions) was used to extract additional structured data (such as the required repair tool "T8 screwdriver").

[0110] Using the Neo4j graph database, a directed graph consisting of triples was constructed: node type: fault type (e.g., "heat dissipation fault"), fault characteristic (e.g., "temperature fluctuation"), maintenance case (e.g., "Case 20230805"), and laser model (e.g., "IPG-YLR-5000"). Relationship types included "has characteristics" (fault type → fault characteristic), "corresponding case" (fault type → maintenance case), and "applicable equipment" (maintenance case → laser model). Each node stored detailed attributes (e.g., the fault characteristic node contained "parameter name," "threshold," and "duration"; the maintenance case node contained "step number," "spare part number," and "time required").

[0111] By converting scattered manual experience into a structured graph, the utilization rate of historical cases has increased from 20% to 80%. This allows for rapid retrieval of similar cases when diagnosing new faults, reducing trial and error. Graph search (for example, searching for "repair cases applicable to IPG-YLR-5000 with fault characteristics including temperatures > 35°C") outputs the most relevant solutions, improving efficiency by 70% compared to manually searching manuals.

[0112] Exemplarily, generating maintenance suggestions corresponding to the potential fault based on the knowledge graph includes: converting the potential fault output by the time series prediction model into a knowledge graph query statement; matching multiple similar fault cases corresponding to the knowledge graph query statement in the knowledge graph through a graph search algorithm; performing similarity calculation on each of the similar fault cases based on the model and operation history data of the laser to determine a target similar fault case among the multiple similar fault cases; generating maintenance suggestions including fault cause analysis, spare parts replacement suggestions and operation step instructions based on the target similar fault case.

[0113] Convert the potential faults output by the time series prediction model (such as "cooling system abnormality") into a knowledge graph query statement (Cypher language): MATCH (f:fault type {name: "cooling system abnormality"})-[:corresponding case]->(c:repair case) RETURN c, and add the device model condition (such as WHERE c.device_model="TRUMPF TruFlow 1000").

[0114] After returning multiple similar fault cases through a graph search algorithm (such as the shortest path search), the system performs a comprehensive score based on the laser model matching degree (if the applicable model of the case is exactly the same as the current device model, the weight is 0.4), the fault feature similarity (the overlap rate of the abnormal parameter threshold is calculated, the weight is 0.3), and the historical repair success rate (the number of times the problem was solved in the case history application, the weight is 0.3), and selects the target similar case with the highest score.

[0115] Extract fault cause analysis (such as "cooling fan bearing wear"), spare parts replacement recommendations (such as "spare part number SP-0012, brand A"), and operation steps (step-by-step graphic instructions, including safety precautions) from the target case, form structured recommendations (such as JSON format), and push them to the user terminal.

[0116] By combining device models and real-time fault characteristics, the system avoids the problem of generic recommendations not being applicable to specific devices, improving the accuracy of repair procedures by 60%. Novice repair personnel can quickly locate problems using recommended cases, reducing repair time by 40%, especially for complex faults (such as multi-parameter correlation anomalies).

[0117] In some embodiments, a data interaction interface is provided to a preset user terminal, including: the application service layer establishes a real-time communication channel through the WebSocket protocol, which is used to push real-time curves, potential faults and maintenance suggestions corresponding to the key parameter set to the user terminal; the data interaction interface supports augmented reality remote collaboration function, and is configured to superimpose the first-person perspective image corresponding to the laser and the maintenance suggestions.

[0118] The application service layer establishes a persistent connection using the WebSocket protocol, pushing real-time curves of key parameters (such as a power-over-time trend graph updated once per second), potential fault warnings (red pop-up windows with fault confidence levels), and maintenance recommendation summaries (such as "Recommend checking the cooling fan; see the steps for details") to the user terminal (PC web page or mobile app). This supports simultaneous display on multiple terminals, allowing administrators to monitor global device status on the web, while on-site maintenance personnel receive real-time alerts via the mobile app.

[0119] Maintenance personnel wearing AR glasses (such as HoloLens) capture the actual laser. The system uses image recognition to locate the device model, loads the corresponding 3D model, and overlays repair recommendations (e.g., an arrow marks the cooling fan location, and a floating window displays "Replacement steps: 1. Disconnect power; 2. Remove three screws"). Remote experts view the first-person view of the AR glasses via a web interface and annotate key components in real time (e.g., circling areas of abnormal heat generation). This annotation information is then displayed simultaneously within the AR interface, creating a "virtual-real" guidance scenario.

[0120] Real-time curves of key parameters allow users to monitor device status in real time, with early warning notifications delivered within seconds. This reduces fault response time from 10 minutes to 30 seconds. Onsite personnel no longer need to frequently consult manuals, and maintenance procedures are intuitively visualized. This reduces the error rate for complex operations (such as internal component replacement) by 50%, and increases remote collaboration efficiency by more than three times.

[0121] In some embodiments, to address the problems of low efficiency in goods entry and exit and prone to errors in inventory counting in small and medium-sized warehouses, an intelligent shelf management system that combines RFID tags and visual recognition technology is designed to achieve automatic positioning of goods, real-time synchronization of entry and exit records, and abnormal status warnings.

[0122] Hardware layer: RFID readers are installed on the shelves (deployed at intervals of 1 meter), and each item is affixed with a passive RFID tag that stores information such as the item ID, specifications, and storage area. An AI camera (with a wide-angle lens and support for 1080P real-time video streaming) is installed on the top of the shelf to identify the placement status of items on the shelf (such as misplacement or vacancies) in real time through the edge computing module.

[0123] Software layer: The central management system connects to the enterprise ERP through API to obtain order information;

[0124] RFID data analysis: Real-time reading of tag signals when goods enter and exit the warehouse, combined with shelf location codes to determine the location of goods; visual recognition algorithm (based on YOLOv8): Training models to identify abnormalities in the appearance and placement of goods, such as damaged labels and tilted goods, and generate early warning information; synchronization of RFID and visual data through timestamps to cross-verify the status of goods (for example, "tag signal disappears but visual detection of residual goods" triggers an abnormality alarm).

[0125] Inbound: Workers place goods on shelves, RFID readers read the tags and associate them with shelf locations. The visual system confirms compliance and then records the information into the system. Outbound: When an order is triggered, the system pushes the picking route. When the picker retrieves the goods, the RFID signal disappears. The visual system verifies that the goods have been removed correctly and prompts corrections if any anomalies occur. Inventory: Regularly update inventory through RFID batch scanning, combined with visual recognition to complete the information on goods with missed or damaged labels. This allows immediate detection of problems such as damaged labels and misplaced goods, preventing subsequent process errors.

[0126] In some embodiments, an environmental monitoring system combining IoT sensor networks and digital twin technology is designed to address the impact of the storage environment (temperature, humidity, smoke, and cargo stacking pressure) on precision electronic components, enabling real-time risk warning and energy consumption optimization.

[0127] Deploy multiple types of sensors: temperature and humidity sensors (accuracy ±0.5°C / ±2%RH), pressure sensors (installed on shelf load-bearing beams to monitor cargo weight distribution), and smoke sensors (to connect to the fire protection system). Edge gateways collect sensor data and upload it to the cloud via the 5G network. A digital twin engine is also deployed locally.

[0128] The digital twin model is a virtual warehouse space constructed using 3D modeling tools. Each shelf / item corresponds to a virtual entity and synchronizes sensor data (such as temperature and humidity changes, and pressure values) in real time.

[0129] When the temperature and humidity exceed the storage standards for electronic components (e.g., temperature > 28°C, humidity > 60%RH), a multi-level alarm (APP push, on-site buzzer) is triggered. The system uses shelf pressure data to identify the risk of cargo stacking overload (e.g., the load capacity of a single shelf exceeds 80% of the design value) and generates cargo relocation suggestions. Combining temperature and humidity data with warehouse personnel movement patterns, the system automatically adjusts the power of air conditioning / ventilation equipment, and activates energy-saving mode during nighttime hours when no one is around.

[0130] The response time for environmental anomalies has been shortened from 30 minutes for manual inspections to real-time warnings, reducing the risk of cargo damage; through intelligent regulation, warehouse energy consumption has been reduced by 25% year-on-year; the digital twin interface presents the warehouse status in real time, and managers can remotely simulate cargo scheduling to optimize storage space utilization.

[0131] In some embodiments, to address the bottleneck in sorting efficiency during peak order periods in e-commerce warehouses, a sorting system combining an AGV robot cluster and a dynamic path planning algorithm is designed to achieve multi-robot collaborative operation and real-time obstacle avoidance.

[0132] AGV robots (equipped with robotic arms or sorting trays) integrate lidar (navigation accuracy ±10mm) and visual recognition cameras (to read cargo manifest information). UWB positioning base stations (millimeter-level positioning) are deployed on the warehouse floor, and millimeter-wave radars are installed in key corridors to detect people and obstacles. After receiving an order, AGV tasks are assigned based on the size of the goods and the location of the sorting port. The Dijkstra algorithm is used to dynamically plan routes to avoid congested areas.

[0133] Location information is shared through real-time communication, and the velocity obstacle method (Velocity Obstacle) is used to avoid collisions. Sorting tasks are dynamically adjusted according to the AGV's battery level and task completion progress to avoid overloading of a single machine. When a person is detected entering the work area, the AGV automatically slows down and switches to safety mode, and resumes normal speed after the person leaves.

[0134] The embodiment of the present application provides a cloud service-based laser data acquisition and management method, which is applied to a cloud service-based laser data acquisition and management system as provided in any embodiment of the present application. Specifically, Figure 2 As shown, the provided cloud service-based laser data collection and management method includes steps S101 to S105. The details are as follows:

[0135] Step S101. Connect multiple lasers separately through the device connection layer to obtain the device fingerprint corresponding to each laser. The device fingerprint includes identification information and instruction set characteristics. Dynamically match the preset protocol library according to the device fingerprint to convert the communication protocol of each laser into a unified data format.

[0136] Specifically, access adaptation of heterogeneous lasers is achieved through the device connection layer. The core technologies include device fingerprint recognition, dynamic protocol matching and data format standardization.

[0137] Device fingerprint: Contains unique identification information (such as device MAC address, serial number, hardware version) and instruction set characteristics (such as communication protocol type, data frame format, and instruction interaction logic) to distinguish communication differences between lasers from different manufacturers and models. Pre-built protocol library: Built-in parsers for mainstream laser communication protocols (such as Modbus TCP, RS-485 custom protocols, OPC UA, etc.), each with corresponding data parsing rules and format conversion templates. Dynamic matching mechanism: Based on the instruction set characteristics in the device fingerprint, a matching protocol parser is automatically selected through a rule engine or machine learning model (such as a decision tree) to convert raw binary / custom format data into a unified JSON format (including device ID, timestamp, and raw data field mapping).

[0138] An industrial-grade gateway (supporting RS-232 / 485, Ethernet, WiFi, and other interfaces) is deployed at the device connection layer to directly connect to the laser through a hardware interface. The gateway has a built-in device discovery module that actively scans for lasers connected to the network and reads the underlying BIOS / firmware information of the device to generate a device fingerprint.

[0139] Supports manual addition / update of protocol parsing plug-ins (such as Python scripts), which can take effect without restarting the system through hot loading technology;

[0140] Raw data → Protocol parser decoding → Field mapping (for example, mapping the "0x0103" instruction to the power parameter) → Unified format encapsulation (for example: {"device_id":"L-001","timestamp":"20231001","power":35.2,"temp":28.5}).

[0141] It supports mixed access of lasers from multiple manufacturers and models (compatibility rate ≥ 95%), solving the pain point of "one manufacturer, one adapter" in traditional systems. It automatically completes protocol matching when devices are connected, reducing manual configuration costs and shortening deployment time from 2 hours / device in traditional solutions to 5 minutes. The unified format lays the foundation for subsequent data analysis, avoiding data processing errors caused by protocol differences.

[0142] Step S102. Through the edge computing preprocessing module deployed at the local end of the laser, the original collected data of the laser is filtered, compressed and feature extracted to generate a key parameter set including power, temperature and operating status.

[0143] Specifically, by utilizing the edge computing module to complete noise reduction filtering, compression optimization, and feature extraction of the raw data at the local end of the laser, the computing pressure on the cloud is reduced.

[0144] Filtering processing uses digital filtering algorithms (such as Kalman filtering and mean filtering) to remove sensor noise (such as high-frequency jitter in power data). The compression algorithm retains key feature points by performing lossy / lossless compression (such as ZSTD and GZip) on the original waveform data (such as current waveform and temperature curve). Feature extraction uses signal processing technology (such as Fourier transform and wavelet transform) to extract key parameters such as power stability, temperature fluctuation amplitude, and operating status code (normal / warning / fault) from the original time domain / frequency domain data.

[0145] Deploy a lightweight edge computing box, configure it with an ARM processor and local storage, and connect it to the laser sensor through GPIO / network port. It supports offline caching function, temporarily storing data when the network is interrupted and automatically resuming the data after recovery.

[0146] The filtering module presets noise thresholds for different laser types (such as the allowable temperature fluctuation range of ±1°C for high-power lasers). Exceeding the limit data triggers secondary filtering; for example, the effective value, peak value, and harmonic components are extracted from the current waveform to generate a "power stability index" (0-100, with higher values indicating greater stability).

[0147] The original data volume compression rate reaches 70%-80%, reducing cloud storage costs and network transmission delays (bandwidth usage is reduced by 60%); key parameters are generated locally in real time, responding to device status changes in milliseconds, providing a high-quality data source for subsequent fault prediction; the edge side undertakes 80% of the preprocessing calculations, allowing cloud resources to focus on complex model operations.

[0148] Step S103: Through the access service layer, a load balancer and a data receiving queue are used to perform distributed access processing on the key parameter set.

[0149] Specifically, the access service layer achieves reliable access to high-concurrency data through load balancers and data receiving queues. The core technologies include traffic distribution, queue peak shaving and valley filling, and distributed transaction processing.

[0150] Use polling, minimum number of connections, source IP hashing and other algorithms to evenly distribute access requests to multiple data receiving nodes; use distributed queue systems (such as Kafka and RabbitMQ) to cache pending data, support message persistence and retry mechanisms to avoid data loss; deploy stateless service instances and support dynamic scaling (such as Alibaba Cloud ECS elastic scaling) to cope with scenarios where the number of lasers changes dynamically.

[0151] The load balancing layer uses Nginx or a cloud-native load balancer (such as AWS ALB), listens to a unified access port (such as 8080), and configures a health check mechanism (checking node availability every 5 seconds). The queue system creates topics (Topics) based on laser type / region, such as "laser-east-power" and "laser-west-status", and supports consumer grouping and parallel processing.

[0152] The queue configuration uses the acks=all mechanism to ensure that messages are consumed at least once. Unacknowledged messages are requeued after 30 seconds to prevent data loss caused by network fluctuations. A single cluster can support the simultaneous access of over 100,000 lasers, with a peak throughput of 50MB / s, meeting the needs of large-scale industrial scenarios. Queuing reduces processing latency for burst traffic (such as batch device rollouts) from seconds to milliseconds, achieving 99.95% system availability. Dynamic, on-demand expansion of access nodes improves resource utilization by 40%, eliminating the resource waste associated with traditional fixed architectures.

[0153] Step S104. Through the core processing layer, receive the key parameter set and the environmental parameters corresponding to the laser, and input them into the timing prediction model built based on the long short-term memory network, and output the potential fault corresponding to the laser; by constructing a knowledge graph including preset fault characteristics and maintenance cases, generate maintenance suggestions corresponding to the potential fault according to the knowledge graph.

[0154] Specifically, the core processing layer combines the LSTM time series prediction model and the fault knowledge graph to achieve intelligent diagnosis, including data fusion, model prediction and maintenance suggestion generation.

[0155] The time series prediction model inputs key parameters (power, temperature, operating time) and environmental parameters (ambient temperature and humidity, grid voltage), uses an LSTM network to capture long-term device status change trends, and outputs potential failure probabilities (e.g., "laser cavity aging probability 85%").

[0156] Knowledge graph construction integrates historical fault data (such as fault code, occurrence time, damaged parts) and repair cases (such as replacement steps and spare part models) to build a "fault characteristics-repair strategy" association graph, supporting fast queries based on graph databases (such as Cypher language retrieval);

[0157] The decision engine is used to match the model prediction results with the fault characteristics in the knowledge graph to generate specific maintenance suggestions (such as "Temperature continues to exceed the limit → It is recommended to check the cooling fan, spare part code: SP-003").

[0158] Data preprocessing involves normalizing the time series data (Min-Max scaling), splitting the training and test sets into a 7:3 ratio, and generating samples using a sliding window (window size = 24 hours of data). The LSTM parameters include three hidden layers, 128 neurons per layer, and the Adam optimizer is used. The loss function is cross-entropy, and the training cycle is 200 rounds. The validation set accuracy is ≥92%.

[0159] Knowledge graph construction: Entity definition: including entity types such as "laser model", "fault type", "repair steps", and "spare parts"; relationship extraction: parsing historical maintenance reports through natural language processing (NLP), automatically extracting relationships such as "fault-cause" and "cause-solution", and importing them into the Neo4j graph database after manual verification.

[0160] Potential faults (such as cavity contamination and pump source attenuation) can be predicted 24-48 hours in advance, discovering hidden problems earlier than traditional threshold alarms; the accuracy of maintenance recommendations is over 85%, and the average fault handling time is shortened from 4 hours to 1.5 hours, reducing equipment downtime losses; through the accumulation of enterprise-specific maintenance experience through knowledge graphs, the training cycle for new engineers is shortened by 50%, avoiding the problem of "experience loss".

[0161] Step S105: Provide a data interaction interface to a preset user terminal through the application service layer, where the user terminal includes a smart device that supports augmented reality remote collaboration to receive maintenance suggestions.

[0162] Specifically, the application service layer provides a multi-terminal data interaction interface, the core of which includes RESTful API services and AR remote collaboration functions, supporting users to obtain equipment status and maintenance instructions in real time.

[0163] The data interaction interface provides APIs such as equipment monitoring (real-time parameter query), report download (monthly operation report), and early warning subscription (email / SMS notification), and supports OAuth2.0 authentication. It also enables remote maintenance guidance based on smart glasses, transmitting on-site images in real time through video streaming, superimposing virtual annotations (such as "Here is the temperature sensor installation location") and maintenance step instructions.

[0164] An API gateway was built using the Spring Boot framework, defining standardized interface documentation (Swagger format) and supporting JSON / Protobuf data formats. iOS / Android apps were developed, enabling real-time data push via WebSocket (e.g., notifying users within 1 second of device status changes). The Vuforia SDK recognized the laser body logo and automatically linked the device documentation and maintenance manual. On-site personnel wearing AR devices initiated collaboration requests, which the expert received via the video stream. The expert then circled the faulty component using annotation tools, which then displayed animated repair steps. Users could access the system through multiple devices, including mobile phones, PCs, and AR devices, to monitor device status in real time, with response times reduced to seconds. AR collaboration reduced reliance on on-site engineers, resulting in a 70% remote resolution rate for complex faults and a 60% reduction in travel costs. By overlaying information with AR, understanding repair steps was reduced by 40%, making it particularly suitable for maintenance scenarios involving complex optical paths and circuits.

[0165] It should be noted that technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the above-described cloud service-based laser data acquisition and management method and the specific working process of each step can refer to the corresponding process in the cloud service-based laser data acquisition and management system embodiments described in the above-mentioned embodiments, and will not be repeated here.

[0166] See also Figure 3 As shown, Figure 3 2 is a schematic diagram of the structure of a cloud-based laser data acquisition and management device 200 provided in an embodiment of the present application. This cloud-based laser data acquisition and management device 200 is used to execute the steps of the cloud-based laser data acquisition and management method described in each of the above embodiments. This cloud-based laser data acquisition and management device 200 can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0167] like Figure 3 As shown, the laser data acquisition and management device 200 based on cloud service includes:

[0168] The protocol conversion unit 201 is configured to connect to multiple lasers through a device connection layer, obtain a device fingerprint corresponding to each laser, the device fingerprint including identification information and instruction set characteristics, and dynamically match a preset protocol library based on the device fingerprint to convert the communication protocol of each laser into a unified data format;

[0169] The parameter generation unit 202 is configured to filter, compress, and extract features from the original collected data of the laser through an edge computing preprocessing module deployed at the local end of the laser to generate a key parameter set including power, temperature, and operating status;

[0170] The service access unit 203 is configured to perform distributed access processing on the key parameter set by using a load balancer and a data receiving queue through an access service layer;

[0171] The suggestion generating unit 204 is configured to receive the key parameter set and the environmental parameters corresponding to the laser through the core processing layer, input them into a time series prediction model constructed based on a long short-term memory network, and output a potential fault corresponding to the laser; construct a knowledge graph including preset fault characteristics and maintenance cases, and generate a maintenance suggestion corresponding to the potential fault based on the knowledge graph;

[0172] The interaction completion unit 205 is configured to provide a data interaction interface to a preset user terminal through an application service layer, wherein the user terminal includes a smart device supporting augmented reality remote collaboration, so as to receive the maintenance suggestion.

[0173] It should be noted that technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working processes of the cloud service-based laser data acquisition and management device and each module described above can refer to the corresponding processes in the cloud service-based laser data acquisition and management method embodiments described in the above embodiments, and will not be repeated here.

[0174] The above-mentioned laser data acquisition and management method based on cloud service can be implemented in the form of a computer program. The computer program can be used in Figure 3 Run on the device shown.

[0175] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the steps of the cloud service-based laser data acquisition and management method provided in any embodiment of the present application.

[0176] The computer-readable storage medium may be an internal storage unit of the control module described in the aforementioned embodiment, such as a hard disk or memory of the control module. The computer-readable storage medium may also be an external storage device of the control module, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the control module.

[0177] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A laser data acquisition and management system based on cloud services, characterized in that: include: The device connection layer is configured to connect to multiple lasers and dynamically match a preset protocol library based on the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format; the device fingerprint includes identification information and instruction set characteristics; An edge computing preprocessing module, deployed on the local side of the laser, is used to filter, compress, and extract features from the raw collected data of the laser to generate a set of key parameters including power, temperature, and operating status; an access service layer configured to perform distributed access processing on the key parameter set through a load balancer and a data receiving queue; The core processing layer receives the key parameter set and environmental parameters corresponding to the laser and inputs them into a time series prediction model built based on a long short-term memory network to output potential faults corresponding to the laser; constructs a knowledge graph including preset fault characteristics and maintenance cases to generate maintenance suggestions corresponding to the potential faults based on the knowledge graph; The application service layer is used to provide a data interaction interface to a preset user terminal, and the user terminal includes a smart device that supports augmented reality remote collaboration and is used to receive the maintenance suggestion.

2. The system according to claim 1, wherein: The dynamically matching a preset protocol library according to the device fingerprint corresponding to the laser to convert the communication protocol of each laser into a unified data format includes: parsing identification information and instruction set features in the device fingerprint; The corresponding protocol adapter is called from the preset protocol library according to the identification information and instruction set characteristics; the protocol adapter is used to convert the communication protocol of each laser into the unified data format.

3. The system according to claim 1, wherein: The raw data collected from the laser is filtered, compressed, and feature extracted to generate a set of key parameters including power, temperature, and operating status, including: According to the preset digital filtering algorithm, the noise signal in the original collected data is processed by mean filtering or median filtering; Using a lossy compression algorithm to compress redundant data in the original collected data and retain valid data points; The power fluctuation peak value and temperature change rate of the original collected data are extracted through time domain feature analysis, and are associated with the operating status identifier corresponding to the original data to generate the key parameter set.

4. The system according to claim 1, wherein: The environmental parameters include temperature and humidity; before the input into the time series prediction model constructed based on the long short-term memory network, it also includes: A time series prediction model is constructed based on training of historical data corresponding to the laser; during the training process of the time series prediction model, a sliding window method is used to segment the time series data corresponding to the historical data; the time series prediction model is used to capture the long-term dependency between key parameter sets and environmental parameters through long-short-term memory network units, and output potential faults including component life prediction of the laser and abnormal state warning.

5. The system according to claim 1, wherein: The distributed access processing of the key parameter set by the load balancer and the data receiving queue includes: The load balancer is configured to distribute access requests to multiple parallel data processing nodes based on a polling strategy or a load weight strategy; The data receiving queue adopts a distributed message queue architecture to buffer and control the flow of the key parameter set, ensuring that the key parameter set is connected to the core processing layer at a preset time interval.

6. The system according to claim 1, wherein: The construction includes a knowledge graph of preset fault characteristics and maintenance cases, including: Using natural language processing technology, entity recognition and relationship extraction are performed on preset historical fault reports and maintenance manuals to extract the correlation between fault characteristics and maintenance cases; the fault characteristics include fault codes and abnormal parameter thresholds, and the maintenance cases include maintenance steps and laser models; A graph database is used to store the fault characteristics and maintenance cases, and a directed graph structure containing triplets of fault type, cause, and solution is constructed to form the knowledge graph.

7. The system according to claim 6, characterized in that Generating a maintenance suggestion corresponding to the potential fault according to the knowledge graph includes: Converting the potential faults output by the time series prediction model into a knowledge graph query statement; Matching multiple similar fault cases corresponding to the knowledge graph query statement in the knowledge graph through a graph search algorithm; performing similarity calculation on each of the similar fault cases according to the model and operation history data of the laser, so as to determine a target similar fault case from the plurality of similar fault cases; Generate maintenance recommendations based on the target similar failure cases, including failure cause analysis, spare parts replacement suggestions, and operation procedure guidance.

8. The system according to claim 1, wherein: Provides a data interaction interface to the preset user terminal, including: The application service layer establishes a real-time communication channel through the WebSocket protocol, which is used to push real-time curves, potential faults and maintenance suggestions corresponding to the key parameter set to the user terminal; the data interaction interface supports augmented reality remote collaboration function, and is configured to superimpose the first-person perspective image corresponding to the laser and the maintenance suggestions.

9. A laser data collection and management method based on cloud services, characterized in that: The method applied to the cloud-based laser data acquisition and management system according to any one of claims 1 to 8 comprises: Connecting multiple lasers through a device connection layer, obtaining a device fingerprint corresponding to each laser, wherein the device fingerprint includes identification information and instruction set characteristics, and dynamically matching a preset protocol library according to the device fingerprint to convert the communication protocol of each laser into a unified data format; The edge computing pre-processing module deployed at the local end of the laser filters, compresses and extracts features from the original collected data of the laser to generate a key parameter set including power, temperature and operating status; Through the access service layer, a load balancer and a data receiving queue are used to perform distributed access processing on the key parameter set; The core processing layer receives the key parameter set and environmental parameters corresponding to the laser, inputs them into a time series prediction model built based on a long short-term memory network, and outputs potential faults corresponding to the laser; constructs a knowledge graph including preset fault characteristics and maintenance cases, and generates maintenance suggestions corresponding to the potential faults based on the knowledge graph; A data interaction interface is provided to a preset user terminal through an application service layer, and the user terminal includes a smart device that supports augmented reality remote collaboration to receive the maintenance suggestion.

10. A laser data acquisition and management device based on cloud services, characterized in that: The laser data acquisition and management system based on cloud services according to any one of claims 1 to 8, wherein the device comprises: A protocol conversion unit is configured to connect to multiple lasers through a device connection layer, obtain a device fingerprint corresponding to each laser, the device fingerprint including identification information and instruction set characteristics, and dynamically match a preset protocol library based on the device fingerprint to convert the communication protocol of each laser into a unified data format; A parameter generation unit is used to filter, compress and extract features from the original collected data of the laser through an edge computing preprocessing module deployed at the local end of the laser to generate a key parameter set including power, temperature and operating status; A service access unit, configured to perform distributed access processing on the key parameter set by using a load balancer and a data receiving queue through an access service layer; a suggestion generating unit, configured to receive, through the core processing layer, the key parameter set and environmental parameters corresponding to the laser, input them into a time series prediction model constructed based on a long short-term memory network, and output a potential fault corresponding to the laser; construct a knowledge graph including preset fault characteristics and repair cases, and generate a repair suggestion corresponding to the potential fault based on the knowledge graph; The interaction completion unit is used to provide a data interaction interface to a preset user terminal through an application service layer, and the user terminal includes a smart device that supports augmented reality remote collaboration to receive the maintenance suggestion.

Images