A digital-twin-based full-life-cycle management system for a hand drill

Abstract

The present application relates to the technical field of industrial internet of things and equipment health management, in particular to a full life cycle management system for a hand drill based on digital twinning, comprising a data acquisition module for acquiring physical signals reflecting operating conditions from the hand drill body; an end-side processing module for processing the physical signals to identify life cycle events and generate semantic event messages; a digital twinning module for updating a digital twin representing the health status of the hand drill according to the semantic event messages; an optimization module for correlating the semantic event messages with physical failure modes to feed back system parameters in a closed-loop manner; and an application module for calculating the remaining useful life and providing predictive maintenance decisions based on the digital twin. The system realizes a communication mode transformation from data-driven to event-driven, significantly reducing the communication data volume, device power consumption and cloud computing load, and providing a technical prerequisite for the access and management of large-scale equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial Internet of Things and equipment health management, in particular to a full life cycle management system for a hand drill based on digital twinning. BACKGROUND

[0002] In the current field of industrial equipment management, higher and higher requirements are put forward for the reliability and maintenance efficiency of electric tools such as hand drills. In order to ensure that the equipment can operate stably and reduce the loss caused by unplanned downtime during its entire life cycle, enterprises usually need to continuously monitor the running state and evaluate the health condition and predict the remaining life by analyzing the running data; traditional equipment management methods mostly rely on periodic preventive maintenance or responsive maintenance after the fact, while modern predictive maintenance solutions tend to upload massive raw sensor data generated during equipment operation to the cloud for centralized analysis;

[0003] In the prior art, cloud-based centralized analysis solutions can process equipment data, but they generally rely heavily on high-bandwidth raw data streams; this architecture requires the equipment to continuously transmit high-frequency collected raw time series data such as vibration, current, and temperature to the cloud server, resulting in extremely high communication power consumption and network bandwidth costs, and also bringing huge computing and storage pressure to the cloud, which poses technical and economic obstacles to the access and management of large-scale equipment; in addition, existing life assessment models mostly rely on single-dimensional indicators such as cumulative running time, and the quantization accuracy of nonlinear damage accumulation caused by different load conditions is insufficient, resulting in deviations between the health status assessment results and the physical reality;

[0004] More importantly, the analysis models of traditional predictive maintenance systems are usually static; once these models are deployed, their internal parameters remain unchanged, and there is no effective feedback mechanism to use real equipment failure data to correct and optimize the models; over time, the error between the model and the actual physical degradation process of the equipment will gradually accumulate, reducing the accuracy of the prediction and the effectiveness of the maintenance decision. Therefore, how to provide a full life cycle management system for a hand drill that can realize end-side intelligent processing, accurate state representation, and closed-loop self-optimization is a technical problem that technicians in the field need to solve. SUMMARY

[0005] To solve the above technical problems, the present application discloses a full life cycle management system for a hand drill based on digital twinning, in particular, the technical solution comprises:

[0006] A full life cycle management system for a hand drill based on digital twinning comprises:

[0007] A data acquisition module for acquiring physical signals reflecting the running conditions from the hand drill body;

[0008] an end-side processing module configured to process the physical signals to identify life cycle events and generate semantic event messages;

[0009] a digital twin module configured to update a digital twin representing the health state of the electric hammer drill based on the semantic event messages;

[0010] an optimization module configured to correlate the semantic event messages with physical failure modes to feedback optimize system parameters in a closed loop manner;

[0011] an application module configured to calculate remaining useful life and provide predictive maintenance decisions based on the digital twin.

[0012] Preferably, the end-side processing module is specifically configured to:

[0013] perform time domain analysis and frequency domain analysis on the discrete time series of the physical signals;

[0014] extract a multi-dimensional feature vector within a sliding time window;

[0015] the physical signals include body vibration signals, motor driving current signals, and motor temperature signals.

[0016] Preferably, the end-side processing module is further specifically configured to:

[0017] perform deep fusion on the time series of the multi-dimensional feature vector using a transformer model based on an attention mechanism;

[0018] generate a low-dimensional semantic state vector;

[0019] the semantic state vector is used to capture the internal correlation of complex working conditions.

[0020] Preferably, the end-side processing module is further specifically configured to:

[0021] input the semantic state vector into an in-machine event classifier to obtain a probability distribution vector corresponding to a predefined event set;

[0022] compare a probability component corresponding to an event in the probability distribution vector with a preset confidence threshold;

[0023] when the probability component exceeds the confidence threshold, it is determined that the event occurs;

[0024] when the probability component is not greater than the confidence threshold, it is determined that the event does not occur;

[0025] when it is determined that the event occurs, a semantic event message containing the device identifier and the event type is generated.

[0026] Preferably, the digital twin module is specifically configured to:

[0027] A health status vector for the electric drill;

[0028] The health status vector consists of components that characterize the remaining durability of critical components;

[0029] When a semantic event message is received, the impairment function corresponding to the event type is invoked;

[0030] Calculate the damage increment vector;

[0031] Update the health status vector based on the difference between the health status vector and the damage increment vector.

[0032] Preferably, the non-zero damage increment in the damage increment vector is calculated based on a nonlinear fatigue accumulation model;

[0033] Incremental damage per incident for gearbox components The calculation formula is:

[0034]

[0035] in, The peak current of the event. Rated current, For lossless reference current, For the duration of the event, The damage rate coefficient, For load index, This is a time index.

[0036] Preferably, the application module is specifically used for:

[0037] Obtain the current remaining durability and initial rated life of each key component;

[0038] Determine the current remaining lifespan based on the barrel principle. ;

[0039] The formula for calculating the current remaining useful life is:

[0040]

[0041] in, This represents the total number of critical components. For the first The current remaining durability of each key component. For the first The initial rated life of each key component.

[0042] Preferably, the optimization module is specifically used for: constructing a knowledge graph of the physical failure modes of associated equipment and the life cycle event message sequence; training a graph neural network model based on the knowledge graph; using the graph neural network model to predict the probability that a given event sequence will lead to a specific failure mode; updating system parameters by minimizing the loss function between the predicted failure probability and the actual observation result; and deploying the updated system parameters through over-the-air download technology.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. This system processes data at the end of the electric drill equipment, purifying the high-bandwidth raw physical signals into low-bandwidth semantic event information, realizing a shift from data-driven to event-driven communication mode, significantly reducing the amount of communication data, equipment power consumption and cloud computing load, and providing a technical prerequisite for the access and management of large-scale equipment.

[0045] 2. This system constructs a digital twin and uses a nonlinear fatigue accumulation model to quantify and analyze the damage caused to key components by various operational events. It transforms the fuzzy concept of equipment wear into a traceable and quantifiable multi-dimensional health status evolution process, realizing a precise dynamic characterization of equipment health status and laying a model foundation for accurate life prediction.

[0046] 3. This system, through its application module, transforms the quantified health status information in the digital twin into a remaining service life prediction that users can directly understand, based on the principle that bottleneck components determine the overall lifespan. It can also identify key bottleneck components that determine the lifespan of equipment, making predictive maintenance decisions more targeted and effectively avoiding unplanned downtime.

[0047] 4. This system constructs a feedback loop from actual physical failure of equipment to system model parameters through an optimization module. The system can learn from historical failure cases through knowledge graphs and graph neural network models, and backpropagate to optimize key parameters in the damage model, realizing the self-evolution of the model and ensuring the accuracy and reliability of the system in long-term operation. Attached Figure Description

[0048] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0049] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0051] Example 1:

[0052] Referring to Figure 1 A digital-twin-based full life cycle management system for a hand drill, comprising:

[0053] A data acquisition module for acquiring physical signals reflecting operating conditions from the hand drill body;

[0054] An end-side processing module for processing the physical signals to identify life cycle events and generate semantic event messages;

[0055] A digital twin module for updating a digital twin representing the health status of the hand drill according to the semantic event messages;

[0056] An optimization module for correlating the semantic event messages with physical failure modes to feed back system parameters in a closed-loop manner;

[0057] An application module for calculating the remaining useful life and providing predictive maintenance decisions based on the digital twin;

[0058] A digital-twin-based full life cycle management system for a hand drill, which aims to realize precise, dynamic and predictive closed-loop management of the entire process of the hand drill from operation to failure through an end-cloud collaborative technical architecture; the system includes a data acquisition module, an end-side processing module, a digital twin module, an optimization module and an application module, which work together to form a complete and self-consistent technical closed loop;

[0059] The data acquisition module is a collection of hardware and firmware responsible for obtaining original physical signals from the hand drill body, and its purpose is to provide the entire system with the most basic physical perception data that can reflect the real-time operating conditions of the hand drill; this module is integrated inside the hand drill body, and its role is to provide real-time data for the digital twin model of the system, which comes from multiple sensors integrated inside the hand drill; specifically, a three-axis micro-electromechanical system accelerometer is used to collect vibration signals of the body; a current sensor based on a shunt resistor is used to collect motor drive current signals; and a thermistor is used to monitor the temperature signals of the motor winding; the original physical signals collected by these sensors are subjected to preliminary amplification and filtering by a signal conditioning circuit, and then converted into digital discrete time series as input for subsequent processing;

[0060] The setting purpose of the end-side processing module is to perform real-time processing on high-dimensional and high-bandwidth raw data at the device end close to the data source to extract low-dimensional, low-bandwidth but high information density key event information; in the embodiment, the module first performs feature extraction on the collected data, performs multi-modal data fusion, and finally performs event recognition and determination based on the fused information; the core value of the module lies in converting the original data-driven communication mode into an efficient event-driven communication mode, which provides a technical premise for realizing large-scale device access and lightweight management;

[0061] The digital twin module is deployed in a cloud server, and its core purpose is to maintain a dynamic virtual digital copy that is real-time synchronized with the entity state for each physical electric drill device; the copy is a digital twin; the digital twin is a virtual mapping of the physical electric drill in the digital space, and its function is to represent the health status of the physical device through a state vector composed of the residual durability of key components, and its source is a software object instantiated and maintained by the cloud server for each device; in the embodiment, the module receives semantic event messages generated from the end-side processing module; each message serves as a driving event to trigger the update of the internal state of the digital twin; specifically, the internal core of the digital twin is a health state vector, and the module deducts the health vector according to the type and carried load information of the event, thereby accurately simulating the wear and aging of the physical device due to operation;

[0062] The setting purpose of the optimization module is to establish a feedback loop between the final failure state of the device in the physical world and the event sequence in the digital world, so that the entire system has the ability of self-learning and self-evolution; in the embodiment, the module collects real physical failure reports of all returned devices for repair or scrapped devices, and associates them with all semantic event message sequences uploaded in the life cycle of the device to construct a huge knowledge graph; based on the graph, the key model parameters in the system are continuously iteratively optimized through deep learning technologies such as graph neural networks;

[0063] The setting purpose of the application module is to convert the technical state information contained in the digital twin module into visualized insights and decision support services facing the end users and having direct business value; in the embodiment, the module dynamically calculates and outputs the predicted remaining useful life of the equipment based on the current health state vector of the digital twin and in combination with the initial rated life of each component; meanwhile, it also provides a fleet-level equipment management dashboard and automatically triggers a predictive maintenance work order when the equipment health degree drops to a warning threshold, guiding the maintenance personnel to intervene in advance; the warning threshold can be preset after comprehensive evaluation according to the criticality of the equipment, the availability of spare parts and the maintenance cost; for example, the warning can be triggered when the remaining durability is less than 30% for a core equipment of the production line, and the warning can be triggered when the remaining durability is less than 15% for a non-core equipment.

[0064] The application builds a complete life cycle management closed loop through the organic combination of the above-mentioned modules; it not only stops at monitoring the current state of the equipment, but also realizes the accurate quantification of the health condition of the electric drill, the reliable prediction of the remaining life and the active optimization of the maintenance strategy through intelligent event recognition at the end side, digital twin state evolution in the cloud, model optimization based on real failure data and finally application layer decision support, thereby significantly improving the reliability of the equipment and reducing the unplanned downtime and the life cycle maintenance cost.

[0065] Embodiment 2:

[0066] The end side processing module is specifically used for: performing time domain analysis and frequency domain analysis on the discrete time sequence of the physical signal; extracting a multi-dimensional feature vector in a sliding time window; the physical signal includes a machine body vibration signal, a motor driving current signal and a motor temperature signal;

[0067] The end side processing module is also specifically used for: adopting a transformer model based on an attention mechanism to deeply fuse the time sequence of the multi-dimensional feature vector; specifically, the transformer model adopts a structure containing only an encoder, learns the dependency weight between each dimension and the time sequence of the feature vector through a multi-layer multi-head self-attention network, and generates a generated low-dimensional semantic state vector with higher information density; the semantic state vector is used to capture the internal correlation of complex working conditions;

[0068] The end side processing module is further specified based on embodiment 1, and its internal processing process is designed as a hierarchical progressive information purification process to realize the conversion from the original signal to the low-dimensional semantic vector;

[0069] The end-side processing module processes the discrete time series output by the data acquisition module, i.e., the fuselage vibration signal, the motor driving current signal, and the motor temperature signal; the purpose is to extract key features that can effectively represent the running state of the equipment in a specific time period from high-dimensional original time series data; within a preset size sliding time window, the module performs time domain analysis and frequency domain analysis on the data in the window in parallel; for example, for the motor driving current signal, the root mean square value is calculated to represent the load size; for the fuselage vibration signal, the energy distribution of a specific frequency band is extracted by applying short-time Fourier transform to identify potential mechanical abnormalities; these calculated statistics, together with the average temperature value within the time window, form a multi-dimensional feature vector; the multi-dimensional feature vector here is a set of numerical values that can quantitatively represent the state of the equipment in a fixed time window by statistical calculation and transformation of multi-source original signals, its role is to convert dynamic time series into static format suitable for machine learning model processing, and its source is the feature extraction algorithm in the end-side processing module;

[0070] To further solve the complex nonlinear coupling relationship between the feature dimensions in the multi-dimensional feature vector and capture the dynamic dependence of the state over time, the end-side processing module further adopts a transformer model based on an attention mechanism to deeply fuse the time series of the multi-dimensional feature vector output by the previous stage; the structure of the transformer model is pre-designed and offline trained, and the core self-attention mechanism can dynamically calculate the correlation weights between different sensor features and between the same feature at different time steps under the current working condition; in this way, the model can automatically focus on the most critical feature combination for current state determination, effectively suppressing noise and irrelevant information interference; the final output of this fusion process is a low-dimensional semantic state vector with significantly reduced dimension but higher information entropy; this vector is a compact and semantic representation of the comprehensive running state of the equipment at a certain time, which is not simply a feature splicing, but a highly abstract and general expression of the internal correlation in the current complex working condition of the equipment, and its source is the output of the deep fusion model based on the attention mechanism.

[0071] Embodiment 3:

[0072] The end-side processing module is further specifically used for:

[0073] inputting the semantic state vector into the in-machine event classifier to obtain a probability distribution vector corresponding to a pre-defined event set;

[0074] comparing the probability component corresponding to the event in the probability distribution vector with a preset confidence threshold;

[0075] when the probability component exceeds the confidence threshold, determining that the event occurs;

[0076] when the probability component is not greater than the confidence threshold, determining that the event does not occur;

[0077] when the event is determined to occur, generating a semantic event message containing the device identifier and the event type

[0078] This embodiment is based on embodiment 2, after the end-side processing module generates the low-dimensional semantic state vector, it further executes a lightweight on-machine decision-making process, which aims to map the abstract semantic vector to discrete events with clear business meaning;

[0079] Specifically, the end-side processing module is embedded with an on-machine event classifier; the classifier is a pre-trained computing model, such as a lightweight multilayer perceptron or support vector machine, whose parameters are obtained through supervised learning in the offline stage containing a large amount of annotated working condition data; at runtime, it takes the low-dimensional semantic state vector output by the fusion layer as input; the output of the classifier is a probability distribution vector, and the dimension of the vector is equal to the size of a pre-defined event set; this pre-defined event set is a list containing all key operating states that need to be concerned, which is pre-set according to the physical working principle of the electric drill, failure mode and effect analysis, and expert knowledge; its role is to provide a closed and explicit target space for event recognition, and its source is the prior knowledge base of domain experts; the set can include events such as normal start, no-load operation, light-load drilling, heavy-load drilling, stall, and overheat alarm;

[0080] After obtaining the probability distribution vector, the module compares each probability component in the vector, i.e. the probability of occurrence of each pre-defined event, with a pre-set confidence threshold for the event; the confidence threshold is a numerical threshold set independently for each pre-defined event, which balances the sensitivity and specificity of event recognition, and its determination method is to perform receiver operating characteristic curve analysis on the validation data set to find the best balance point;

[0081] The decision logic is as follows: when the probability component of an event exceeds its exclusive confidence threshold, it is determined that the event occurs; otherwise, when the probability component is not greater than the confidence threshold, it is determined that the event does not occur; to avoid repeated triggering of the same event in a short period of time due to signal fluctuations, this embodiment also introduces a micro finite state machine, which only confirms the occurrence of an event when the device migrates from one stable state to another different state;

[0082] When a semantic event is identified, the module generates a structured semantic event message immediately; the message is an extremely lightweight data packet, whose content strictly follows a preset format, and at least contains the unique device identifier of the current device, the time stamp accurate to milliseconds, and the code of the identified event type; in addition, the message can optionally be attached with the snapshots of several key physical quantities most relevant to the event, such as the peak current when the stall event occurs, for the cloud to conduct more detailed damage analysis; after generation, the message is published to the digital twin module in the cloud through a low-power wide-area network.

[0083] Embodiment 4:

[0084] The digital twin module is specifically used for:

[0085] maintaining a health state vector for the electric drill; the health state vector is composed of components representing the remaining endurance of key components; when receiving a semantic event message, a damage function corresponding to the event type is called; a damage increment vector is calculated; the health state vector is updated according to the difference between the health state vector and the damage increment vector;

[0086] This embodiment is based on Embodiment 1, and the digital twin module is further specified, and its core function is to serve as a dynamic quantitative representation and evolution simulator of the health state of the physical device;

[0087] The digital twin module maintains a health state vector for each electric drill connected to the network in the cloud database; the health state vector, denoted as , is a multi-dimensional vector , which quantitatively represents the health level of the electric drill as a whole and its internal key components, and its source is the internal state variable of the cloud digital twin module; each component of the vector is defined as the remaining endurance of the th key component, and its value range is between ; when a new device is shipped, all components of its health state vector are 1, indicating that all components are in good condition; when a component decays to 0, it means that the component has completely failed; key components can include gearboxes, motor brushes, bearings, etc.

[0088] When the module receives a semantic event message from the edge side, it analyzes the event type in the message; the module will call a damage function corresponding to the event type in advance according to the event type; the damage function here, denoted as , is a function that takes the event message The physical load parameters carried by the event are mapped to a mathematical model of dimensionless damage increment, which quantifies the wear or fatigue accumulation caused by a specific type of operating event on the key components of the device. The source is an algorithm library established in advance by combining material fatigue theory, empirical model and experimental data;

[0089] The damage function calculates a damage increment vector; the vector has the same dimension as the health state vector, but is usually sparse, i.e. only the components corresponding to the components related to the event type have non-zero values; for example, the overheat alarm event mainly damages the motor, so only the component corresponding to the motor has a non-zero value in the damage increment vector of the event;

[0090] The module updates the health state vector according to the difference between the current health state vector and the calculated damage increment vector; the update rule can be expressed as: wherein is a damage coupling function, which not only considers the direct damage increment of the event, but also considers the influence of the current health state on the rate of new damage accumulation. In the simplest implementation, the function can be a simple difference, i.e. , corresponding to the rule ; but in a more precise model, the function can introduce a health state-related damage aggravation factor to reflect the physical reality that an aging component will produce greater damage under the same load. Through this event-driven, discrete deduction process, the health state vector of the digital twin can dynamically evolve with the actual use of the physical device, thus reflecting the cumulative damage and health degradation process in real time; to ensure the comprehensiveness of the model, in addition to handling discrete damage triggered by high-load events, the system can also run a time-based baseline degradation model in parallel, which performs a small, continuous deduction on the health state vector components of related components according to the cumulative running time, thus more completely depicting the wear and tear process of the device throughout its life cycle.

[0091] Embodiment 5:

[0092] The non-zero damage increment in the damage increment vector is calculated based on a nonlinear fatigue accumulation model;

[0093] The formula for calculating the single damage increment of the gearbox component is:

[0094]

[0095] wherein, is the peak current of the event, is the rated current, is the damage-free reference current, is the event duration,​ is the damage rate coefficient, is the load index, is the time index;

[0096] This embodiment further refines the calculation method of the damage increment vector based on Embodiment 4, especially for the damage assessment of mechanical transmission core components, a more accurate physical model is introduced;

[0097] In this embodiment, the non-zero damage increment in the damage increment vector is explicitly calculated based on a nonlinear fatigue accumulation model; the purpose of this is to more realistically simulate the nonlinear relationship between material fatigue damage and stress size, action time in the physical world, and to avoid the large error brought by the linear model;

[0098] Specifically, taking the gear box assembly of a hand drill as an example, when high load events such as heavy load drilling or stall occur, its single damage increment The calculation formula is defined as:

[0099]

[0100] Each parameter in the formula is defined as follows:

[0101] is the single event dimensionless damage increment for the gear box assembly, its data type is float, and its source is calculated by this formula;

[0102] is the damage rate coefficient, used to convert physical load to dimensionless damage degree; its dimension is the power of time (i.e. ), to ensure that the dimensions on both sides of the equation are consistent; the data type of this parameter is float, and its source is obtained by material fatigue experiment and historical data regression analysis, and can be iteratively updated by the optimization module;

[0103] is the event peak current, representing the maximum load level during the event, its dimension is ampere (A), and its source is parsed from the semantic event message that triggers the calculation;

[0104] is the non-damage reference current, representing the current threshold that will not cause any fatigue damage to the component, its dimension is ampere (A), and its source is preset according to the device design specifications and material characteristics;

[0105] is the rated current, the maximum current that can be continuously operated on the hand drill design, its dimension is ampere (A), and its source is given by the device nameplate or design manual;

[0106] is the event duration, i.e. the time length of the load action, with the unit of second (s), which is parsed from the semantic event message;

[0107] is the load index, which is a dimensionless parameter representing the non-linear sensitivity of the damage to the magnitude of the current stress, with the data type of float, which is pre-set according to the S-N curve characteristics of the material and can be iteratively updated by the optimization module;

[0108] is the time index, which is a dimensionless parameter representing the non-linear sensitivity of the damage to the duration of the load, with the data type of float, which is pre-set according to the creep and fatigue theory and can be iteratively updated by the optimization module;

[0109] The technical motivation of this formula is that the effective stress leading to damage is compared with a normalized reference stress range , forming a dimensionless load level factor; through the indices and , the non-linear acceleration effect of damage accumulation is captured, i.e. higher load and longer duration will increase more damage disproportionately; to ensure the calculation stability and correctness of the physical meaning of the mathematical model under all possible inputs, robustness checking steps are included in the actual algorithm implementation: at system initialization, the parameters are checked to ensure that the rated current must be greater than the undamaged reference current , and at each call of the damage function, the effective stress part leading to damage is redefined as , so that when the peak current is not greater than the undamaged threshold, the damage increment is naturally zero, thus avoiding meaningless or incorrect calculation due to negative base, and ensuring the compliance of the model with the physical common sense.

[0110] Embodiment 6:

[0111] The application module is specifically used for:

[0112] obtaining the current residual endurance and the initial rated life of each key component; determining the current residual service life according to the principle of the weakest link ; the calculation formula of the current residual service life is:

[0113]

[0114] wherein, is the total number of key components, is the current residual endurance of the th key component, is the initial rated life of the initial rated life of each key component;

[0115] The application module is further specified based on embodiment 1, and one of its core tasks is to convert the relatively abstract health state vector maintained in the digital twin module into an index with clear physical meaning that users can directly understand and use for decision-making;

[0116] In order to calculate the current remaining service life of the device, the application module obtains two key data: one is the current remaining endurance component of each key component from the digital twin module ; The other is the initial rated life of each key component read from the device profile database ; The initial rated life of the component in the brand new state under the standard working condition is designed for the total service life, which provides a benchmark reference for the remaining life calculation, and its source is the design specification of the device manufacturer or a large number of reliability test statistical results, usually in hours;

[0117] After obtaining the data, the module determines the current overall remaining service life of the device according to the principle of determining the overall life of the system by the bottleneck component, denoted as ; The core idea of this principle is that the overall life of a system is determined by the key component with the shortest internal life;

[0118] The formula for calculating the current remaining service life is defined as:

[0119] Each parameter in the formula is defined as follows:

[0120] is the current remaining service life of the device at the current time, with the dimension of hours (h), and its source is calculated by this formula;

[0121] is the total number of key components included in the device, with the data type of integer, and its source is preset in the system configuration;

[0122] is the current remaining endurance of the th key component, which is a dimensionless value within the range of , with the data type of floating point, and its source is obtained from the health state vector of the digital twin module in real time;

[0123] is the initial rated life of the th key component, with the dimension of hours (h), and its source is obtained by querying the device profile database;

[0124] This function links the dimensionless health state vector with the life indicator with explicit time metric; every time the health state vector of the digital twin module is updated due to an event, the application module will re-execute this calculation, thus achieving dynamic refresh of the remaining useful life.

[0125] Embodiment 7:

[0126] The optimization module is specifically used for:

[0127] Building a knowledge graph that links the physical failure modes of the equipment with the message sequences of the life cycle events; wherein, specific life cycle events (such as overload, stall) are taken as nodes, and the time sequence relationship between events is taken as edges, to build an event sequence graph; training a graph neural network model based on the knowledge graph; for example, a graph attention network (GAT) model; the graph neural network model is used to predict the probability of a given event sequence leading to a specific failure mode; updating the system parameters by minimizing the loss function between the predicted failure probability and the real observation; deploying the updated system parameters through over-the-air technology;

[0128] This embodiment is based on Embodiment 1, and the optimization module is further specified, and its function is to build a data-driven self-evolution closed loop, so that the system can learn from historical experience and continuously improve the accuracy of its internal model;

[0129] The core of this optimization module is to build a knowledge graph; this process is automated: when a flashlight drill equipment eventually fails and is repaired or directly scrapped, its physical failure mode will be diagnosed and recorded; the optimization module will strongly associate this real failure result with all the event message sequences uploaded during its entire life cycle, forming a knowledge pair of event sequence → failure mode; a large number of such knowledge pairs collectively form a complex network structure, i.e. a knowledge graph; denoted as , is a graph structure data with specific event types as nodes and time sequence between events as directed edges; each node can contain event duration, peak current and other attributes as its initial features; its role is to systematically store and express the failure knowledge learned from a large number of equipment; its source is the continuous collection of real physical failure data and corresponding full life cycle event flow;

[0130] Based on the built knowledge graph, the module trains a graph neural network model, such as a graph attention network (GAT) model; the reason for choosing a graph neural network model is its powerful graph structure data processing capability, which can effectively capture long-range dependencies and complex pattern combinations in event sequences; the training of this graph neural network model is to predict: given a certain event sequence, the probability of the equipment leading to a specific failure mode;

[0131] The training process of the model is a supervised learning process, which aims to continuously adjust the network parameters of the model itself by minimizing the loss function between the predicted failure probability and the real observation; more importantly, the gradient of the loss function is not only used to update the graph neural network model itself, but also further calculates the gradient of the global adjustable parameters of the system through the back propagation algorithm; these global parameters include but are not limited to: the decision threshold of the end-side event classifier, the damage rate coefficient of the damage accumulation model in the digital twin , load index and time index ; in this way, the system can find systematic deviations such as a certain damage model parameter is set too low, resulting in all gear wear cases being predicted too optimistically;

[0132] After calculating the gradient, the system uses optimization algorithms such as gradient descent to update the system parameters; through over-the-air download technology, these optimized new parameters are safely and efficiently deployed back to the firmware of all online end-side devices and the cloud model, thus completing the entire closed-loop feedback; these system parameters not only include the damage model parameters of the cloud digital twin module and the classifier parameters of the end-side processing module, but also may include fine-tuning of the configuration parameters of the data acquisition module to achieve global optimization of system performance;

[0133] The disclosed hand drill full life cycle management system based on digital twin realizes a substantial breakthrough in existing device management technology by building a closed-loop architecture that is end-cloud collaborative, data-driven, and self-evolutionary;

[0134] Compared with the traditional cloud centralized analysis scheme that relies on high-bandwidth raw data flow, the system sets up an end-side processing module to complete the deep purification from physical signals to semantic information at the device end close to the physical data source; this module extracts multi-dimensional feature vectors from multiple physical signals such as body vibration, motor current and temperature within a sliding time window, and innovatively uses a transformer model based on attention mechanism for deep fusion; this can autonomously learn and capture the internal correlation and temporal dependence between different physical signals, generating a low-dimensional semantic state vector with extremely high information density; then, the vector is mapped to discrete semantic event messages by the in-machine event classifier; this fundamental shift from data-driven to event-driven results in a drop in communication data volume by orders of magnitude, greatly reducing the communication power consumption of the device and the computational load of the cloud, providing a technical prerequisite for the economic and efficient access and management of large-scale devices;

[0135] Compared with the prior art that only relies on a single-dimensional indicator such as running time for life evaluation, the digital twin module constructed by the system can more accurately quantify and dynamically evolve the health state of the equipment; the module maintains a health state vector composed of the residual durability of key components for each electric drill; when receiving the semantic event message uploaded by the terminal side, the system calls the damage function matched with the event type, calculates a damage increment vector, and updates the health state vector by difference; in particular, the system uses a nonlinear fatigue accumulation model to calculate the damage increment, which can more realistically reflect the nonlinear accelerated damage caused by complex working conditions such as high load and long duration; this mechanism converts the vague equipment wear concept into a traceable and quantifiable multi-dimensional health state evolution process, laying a solid model foundation for accurate fault diagnosis and life prediction;

[0136] The application module provided by the system converts the abstract digital twin state into decision support with direct business value; it dynamically calculates the current residual service life of the whole machine based on the health state vector and the initial rated life of each component according to the principle of the weakest link; this method not only provides intuitive and operable health indicators, but also accurately identifies the bottleneck component that determines the overall life of the equipment, making the development of predictive maintenance strategies more targeted, so as to maximize the availability of the equipment with minimized maintenance cost, effectively avoiding the loss caused by unplanned downtime;

[0137] More importantly, the system establishes a feedback loop from physical failure to model parameters through the optimization module, giving the system unprecedented self-learning and self-evolution capabilities; the module builds a knowledge graph that associates physical failure modes with life cycle event sequences, and trains a graph neural network model based on this to predict the probability of a specific event sequence leading to a specific failure mode; by minimizing the loss function between the prediction result and the real observation, the system not only optimizes the prediction model, but also updates and optimizes global system parameters such as damage rate coefficients and load indices in reverse; these updated parameters are deployed to all network devices through over-the-air technology; this closed-loop optimization mechanism enables the system to learn from each real failure, continuously correct the deviation between the digital model and the physical reality, and ensure the accuracy, reliability and advancement of the entire system in the long run.

[0138] The above is only a preferred embodiment of the present application, and is not intended to limit the protection scope of the present application; any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0139] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A digital twin-based full lifecycle management system for electric drills, characterized in that, include: The data acquisition module is used to collect physical signals from the electric drill body that reflect its operating conditions. The edge processing module is used to process physical signals to identify lifecycle events and generate semantic event messages; The digital twin module is used to update the digital twin representing the health status of the electric drill based on semantic event messages; The optimization module is used to correlate semantic event messages with physical failure modes and to feed back and optimize system parameters in a closed-loop manner. Application modules are used to calculate remaining useful life and provide predictive maintenance decisions based on digital twins; The end-side processing module is specifically used for: Perform time-domain and frequency-domain analysis on the discrete-time series of physical signals; Extract multidimensional feature vectors within a sliding time window; Physical signals include fuselage vibration signals, motor drive current signals, and motor temperature signals; The end-side processing module is also specifically used for: An attention-based transformer model is used to perform deep fusion of time series of multidimensional feature vectors; Generate low-dimensional semantic state vectors; Semantic state vectors are used to capture the inherent relationships in complex operating conditions; The end-side processing module is also specifically used for: The semantic state vector is input into the in-machine event classifier to obtain the probability distribution vector of the corresponding predefined event set; The probability components corresponding to the event in the probability distribution vector are compared with a preset confidence threshold. When the probability component exceeds the confidence threshold, the event is determined to have occurred. If the probability component is not greater than the confidence threshold, the event is determined not to have occurred. When an event is determined to have occurred, a semantic event message containing the device identifier and the event type is generated; The digital twin module is specifically used for: A health status vector for the electric drill; The health status vector consists of components that characterize the remaining durability of critical components; When a semantic event message is received, the impairment function corresponding to the event type is invoked; Calculate the damage increment vector; Update the health status vector based on the difference between the health status vector and the damage increment vector.

2. The full lifecycle management system for a digital twin-based electric drill according to claim 1, characterized in that, The non-zero damage increment in the damage increment vector is calculated based on a nonlinear fatigue accumulation model. Incremental damage per incident for gearbox components The calculation formula is: in, The peak current of the event. Rated current, For lossless reference current, For the duration of the event, The damage rate coefficient, For load index, This is a time index.

3. The full lifecycle management system for a digital twin-based electric drill according to claim 1, characterized in that, The application module is specifically used for: Obtain the current remaining durability and initial rated life of each key component; Determine the current remaining lifespan based on the barrel principle. ; The formula for calculating the current remaining useful life is: in, This represents the total number of critical components. For the first The current remaining durability of each key component. For the first The initial rated life of each key component.

4. The full lifecycle management system for a digital twin-based electric drill according to claim 1, characterized in that, The optimization module is specifically used for: constructing a knowledge graph of physical failure modes of associated equipment and lifecycle event message sequences; training a graph neural network model based on the knowledge graph; using the graph neural network model to predict the probability that a given event sequence will lead to a specific failure mode; updating system parameters by minimizing the loss function between the predicted failure probability and the actual observation results; and deploying the updated system parameters through over-the-air download technology.

Images