Intelligent equipment monitoring system with fault prediction and health management function

Abstract

The present application relates to the technical field of intelligent equipment monitoring and health management, in particular to an intelligent equipment monitoring system with fault prediction and health management function, comprising an intelligent equipment, an instruction data acquisition terminal, an operation feedback acquisition terminal, a monitoring server and a management terminal; the monitoring server generates ideal dynamics benchmark data based on ideal dynamics model and control instruction data, extracts degradation mechanism parameters from a degradation mechanism library to generate theoretical abnormal response data, and respectively constructs actual residual data and theoretical residual data, and determines health state results through similarity check; the management terminal receives and displays the health state results, outputs maintenance prompt information, and sends control adjustment information to the controller; the present application can separate working condition changes and fault degradation, and realize early fault prediction and health management under variable working condition.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment monitoring and health management technology, specifically to an intelligent equipment monitoring system with fault prediction and health management functions. Background Technology

[0002] Intelligent equipment operates under multiple working conditions and continuously in CNC machining, automated production lines, and complex electromechanical systems. During operation, the equipment generates various types of data, such as control commands, servo current, position feedback, torque feedback, and vibration feedback. Utilizing this data to monitor the health status of the equipment, predict faults, and manage maintenance has become an important means to improve operational stability and reduce the risk of downtime.

[0003] With the development of intelligent manufacturing technology, equipment health management has gradually evolved from single alarms to online predictive analysis of degradation processes. However, in application scenarios with large load fluctuations and frequent trajectory switching, existing monitoring methods often rely on fixed thresholds, empirical rules, or historical templates that are detached from the current operating conditions. This makes it difficult to effectively distinguish between normal operating condition fluctuations and real degradation anomalies, which can easily lead to false alarms and missed alarms. As a result, the fault type identification is inaccurate and maintenance decisions are delayed, making it difficult to meet the application needs of complex intelligent equipment for early fault prediction and health management. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent equipment monitoring system with fault prediction and health management functions, and to solve the following technical problems: To avoid interference from common patterns caused by actual process actions and changes in operating conditions, the changes in operating conditions are separated from failure degradation as much as possible. It is also easier to actively simulate and distinguish different degradation mechanisms, thereby enabling early failure prediction under varying operating conditions and supporting proactive intervention in maintenance decisions and health management.

[0005] The objective of this invention can be achieved through the following technical solutions: An intelligent equipment monitoring system with fault prediction and health management functions includes: intelligent equipment, command data acquisition terminal, operation feedback acquisition terminal, monitoring server and management terminal; The intelligent equipment includes a controller, an instruction data acquisition terminal that is connected to the controller to acquire control instruction data and send it to the monitoring server; and an operation feedback acquisition terminal that is used to acquire operation feedback data of the intelligent equipment and send it to the monitoring server. The monitoring server is used to generate ideal dynamics baseline data based on a preset ideal dynamics model and control command data, extract degradation mechanism parameters from a preset degradation mechanism library and generate theoretical abnormal response data by combining the ideal dynamics model and control command data, generate actual residual data based on the operation feedback data and ideal dynamics baseline data under the same time sequence, generate theoretical residual data based on the theoretical abnormal response data and ideal dynamics baseline data, and determine the health status result based on the similarity verification between the actual residual data and the theoretical residual data. The management terminal is used to receive and display health status results, output maintenance prompts, and send control adjustment information to the controller of the intelligent equipment.

[0006] Furthermore, the monitoring server is also used for: Retrieve a pre-defined ideal dynamic model from memory or database; Input the control command data into the ideal dynamic model; Output ideal dynamic benchmark data based on the ideal dynamic model; Among them, the ideal dynamic model is used to characterize the response relationship under the conditions of no friction, no wear, no transmission backlash, and structural stiffness parameters at the reference value.

[0007] Furthermore, the monitoring server is also used for: Retrieve a pre-defined degradation mechanism library from memory or database; Extract degradation mechanism parameters from the degradation mechanism library; Input the degradation mechanism parameters into the ideal dynamic model, or modify the ideal dynamic model based on the degradation mechanism parameters; Theoretical anomaly response data is generated based on the revised model and control command data.

[0008] Furthermore, the degradation mechanism parameters include at least one of the following: friction variation parameters, clearance variation parameters, stiffness variation parameters, and torque constant variation parameters.

[0009] Furthermore, the monitoring server is also used for: Actual residual data is generated based on operational feedback data and ideal dynamic baseline data; Theoretical residual data is generated based on theoretical anomaly response data and ideal dynamic benchmark data; The actual residual data and the theoretical residual data are used as input data for similarity verification.

[0010] Furthermore, the monitoring server is also used for: The time-series similarity between the actual residual data and the theoretical residual data after time alignment is calculated based on the dynamic time warping algorithm, or the similarity between residual feature vectors of the same dimension is calculated based on the multidimensional spatial cosine similarity algorithm. In response to a similarity greater than or equal to a preset threshold, the existence of a degradation mechanism corresponding to the theoretical residual data is determined; If the similarity is less than a preset threshold, it is determined that the degradation mechanism corresponding to the theoretical residual data does not exist; The preset threshold is obtained by offline calibration based on the similarity distribution between historical normal samples and faulty samples.

[0011] Furthermore, the monitoring server is also used for: Given that the degradation mechanism corresponding to the theoretical residual data exists, the fault type, fault stage, and remaining life assessment results are determined based on the similarity results, the changing trend of the degradation mechanism parameters, and the pre-acquired historical degradation trajectory. If it is determined that the degradation mechanism corresponding to the theoretical residual data does not exist, the current state is determined to be a non-fault fluctuation state; The health status results include at least one of the following: fault type, fault stage, remaining life assessment results, and non-fault fluctuation status.

[0012] Furthermore, the management terminal is also used for: When the fault stage is greater than or equal to the warning level set according to the fault stage division rules, a warning message is output; when the fault stage is less than the warning level, no warning message is output. When the remaining life assessment result is less than or equal to the life threshold set according to the maintenance strategy, a maintenance prompt message is output; when the current state is a non-fault fluctuation state, a continue operation message is output.

[0013] Furthermore, the management terminal is also used for: Generate control adjustment information based on health status results; Send control adjustment information to the controller of the intelligent equipment; Among them, the control adjustment information is used to adjust the operating speed, acceleration or load distribution, and the maintenance prompt information is used to prompt adjustments to the maintenance cycle or maintenance plan.

[0014] Furthermore, the control command data includes at least one of PLC control logic data, multi-axis position commands, speed commands, and trajectory commands, and the operation feedback data includes at least one of servo current, encoder position, torque feedback value, and vibration feedback value corresponding to the output response of the ideal dynamic model.

[0015] The beneficial effects of this invention are: 1. This invention generates ideal dynamic reference data based on a preset ideal dynamic model and control command data, and generates actual residual data by subtracting it from the operation feedback data. This mechanism provides a dynamic reference that is strictly aligned with the current operating conditions for anomaly identification, effectively separating operating condition fluctuations from actual degradation, and overcoming the defect of traditional fixed thresholds being susceptible to interference from complex trajectories and load changes, resulting in false alarms. 2. This invention extracts degradation mechanism parameters such as friction, clearance, or stiffness from a degradation mechanism library, combines them with control commands to actively deduce theoretical abnormal response data, and constructs theoretical residual data. This method transforms empirical fault knowledge into calculable parameters, and can intuitively present the deviation patterns that different mechanisms should have under the same control trajectory, solving the problem of inaccurate early fault identification caused by being out of the current command context. 3. The system of this invention adopts dynamic time warping algorithm or multidimensional spatial cosine similarity algorithm to perform temporal morphology or multidimensional similarity verification between actual residual data and theoretical residual data; this overcomes the shortcomings of hard alignment comparison which is easily affected by sampling delay or peak offset, effectively reduces the interference caused by amplitude disturbance and misalignment, and realizes stable extraction and reliable matching of weak degradation features of complex equipment. 4. Based on similarity verification and historical trajectory, the system can accurately determine the fault type, assess the fault stage and extrapolate the remaining lifespan, and output a prompt to continue operation when there is no fault. By combining the warning level and lifespan threshold to output graded maintenance prompts, it not only avoids mis-repairs caused by non-fault fluctuations, but also provides accurate decision support for operation and maintenance personnel and solves the problem of delayed maintenance decisions. 5. The management terminal of this invention can automatically generate control adjustment information based on the assessed health status results, and issue adjustment instructions such as operating speed, acceleration or load distribution to the equipment controller; this closed-loop mechanism from monitoring and diagnosis to active control enables the equipment to operate under control through flexible adjustment in the early stage of failure or when spare parts are not available, effectively avoiding rapid degradation and sudden shutdown failures. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram of the structure of the intelligent equipment monitoring system with fault prediction and health management functions provided in the embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 An intelligent equipment monitoring system with fault prediction and health management functions includes: intelligent equipment, command data acquisition terminal, operation feedback acquisition terminal, monitoring server and management terminal; the intelligent equipment includes a controller; The intelligent equipment includes a controller, and an instruction data acquisition terminal is connected to the controller to acquire control instruction data and send it to the monitoring server. The operation feedback acquisition terminal is used to acquire operation feedback data of intelligent equipment and send it to the monitoring server; The monitoring server is used to generate ideal dynamic baseline data based on a preset ideal dynamic model and control command data, extract degradation mechanism parameters from a preset degradation mechanism library and combine them with the ideal dynamic model and control command data to generate theoretical abnormal response data, generate actual residual data based on the operation feedback data and ideal dynamic baseline data under the same time sequence, generate theoretical residual data based on the theoretical abnormal response data and ideal dynamic baseline data, and determine the health status result based on the similarity verification between the actual residual data and the theoretical residual data. The management terminal is used to receive and display health status results, output maintenance prompts, and send control adjustment information to the controller.

[0020] This embodiment provides a fault prediction and health management mechanism for a five-axis CNC machining center. Specifically, the same machining center is used as an example object for the entire description. The machining center includes a CNC controller, X / Y / Z linear feed axes, A / C rotary axes, and a spindle unit. On a precision impeller machining production line, the equipment performs a mixture of roughing, semi-finishing, and finishing tasks for a long time. The load changes greatly and the trajectory switches frequently, making it suitable as a continuous monitoring scenario under variable working conditions.

[0021] Specifically, the instruction data acquisition terminal is deployed between the controller and the servo bus, or connected to the controller's data interface in a mirror listening manner, to collect control instruction data for the current batch processing task; The control command data may include the currently executed interpolation segment number, the position setpoint, velocity setpoint, acceleration setpoint, and trajectory segment switching identifier for each axis; the operation feedback acquisition terminal is connected to the servo driver, encoder acquisition unit, and vibration acquisition module to synchronously acquire actual current, actual position, torque feedback, and vibration feedback; After receiving the above two types of data, the monitoring server reconstructs the ideal dynamic reference data of the equipment under the condition of no friction, no wear, no transmission clearance, and structural stiffness at the reference value, based on the ideal dynamic model preset for the current model and the current control command data. Candidate degradation mechanism parameters are extracted from the degradation mechanism library, such as increased lead screw friction, increased coupling clearance, decreased bearing stiffness, or decreased motor torque constant. These parameters are then injected into the ideal model to generate theoretical abnormal response data corresponding to the same trajectory and speed command. During the residual construction phase, the server executes two parallel differential links: one subtracts the ideal dynamic baseline data from the running feedback data to obtain the actual residual data; the other subtracts the ideal dynamic baseline data from the theoretical abnormal response data to obtain the theoretical residual data; the server performs similarity verification on the two types of residuals to output the health status result. As a specific data processing example; assuming that a certain processing time window only focuses on three channels: X-axis position, current, and vibration, the ideal dynamic reference data within three consecutive sampling times t1, t2, and t3 are: position reference [10.00, 10.50, 11.00], current reference [2.0, 2.4, 2.8], and vibration reference [0.10, 0.12, 0.11]. The actual operational feedback within the same time window is as follows: position feedback [9.98, 10.46, 10.93], current feedback [2.1, 2.7, 3.3], vibration feedback [0.10, 0.18, 0.25]; then the actual residuals can be simplified as: position residual [-0.02, -0.04, -0.07], current residual [0.1, 0.3, 0.5], vibration residual [0, 0.06, 0.14]; If the parameter combination of increased friction and backlash due to wear of the X-axis lead screw pair is taken from the degradation mechanism library, the theoretical abnormal response generated by the model is: position abnormality [9.99, 10.47, 10.92], current abnormality [2.1, 2.6, 3.2], vibration abnormality [0.10, 0.17, 0.23]; Based on this, the theoretical residuals can be obtained: position residual [-0.01, -0.03, -0.08], current residual [0.1, 0.2, 0.4], and vibration residual [0, 0.05, 0.12]. At this point, even if the actual feedback is affected by the change in cutting depth and causes additional fluctuations, as long as the two sets of residuals maintain a high degree of similarity in the direction of change, the time of peak occurrence, and the coupling relationship of each channel, it can be determined that the degradation mechanism is close to the actual state. After receiving the health status results, the management terminal displays the fault type, degradation stage, remaining life assessment results, or non-fault fluctuation status in the form of a machine tool status panel. It also generates maintenance prompts based on the status, such as checking the X-axis lead screw preload status within 48 hours or indicating that the machine is currently in a non-fault fluctuation state and no shutdown is required. When necessary, the management terminal also sends control adjustment information to the controller, such as reducing the maximum acceleration of the corresponding axis, modifying the trajectory transition smoothing parameters, or reallocating the roughing and finishing task cycles.

[0022] In abnormal situations, when instruction data is missing, feedback data timestamp drift exceeds the allowable range, or the current operating condition exceeds the model's applicable boundary, the server will not directly output a fault conclusion, but will first mark the time window as a segment to be reviewed; if the number of segments to be reviewed is consistently less than the preset number, the previous healthy state will be maintained. If the number of attempts exceeds the preset limit, a prompt will be made to recalibrate the model or verify the sensor link. This can avoid misjudgments caused by factors other than the device itself, such as network jitter or instantaneous frame drops in the encoder. For example, during the impeller finishing stage, the A-axis and C-axis perform high-frequency linkage attitude adjustment, while the X-axis simultaneously completes micro-feed; as the tool enters a local high-hardness region of the material, the spindle load increases instantaneously, and both current and vibration rise. Traditional fixed threshold methods can easily trigger bearing abnormality alarms directly. However, this embodiment first reconstructs the ideal response based on the current tool path and feed command, then divides the response difference into actual residuals, and compares them one by one with various theoretical residuals. Finally, it identifies the problem as early wear of the X-axis lead screw rather than spindle failure, thus providing targeted maintenance prompts without interrupting the overall machine cycle.

[0023] Furthermore, to avoid confusion between control adjustment information and maintenance scheduling suggestions at the execution level, the control adjustment information sent to the controller in this embodiment is preferably limited to parameters or limits that the controller can directly execute, such as single-axis speed limit, single-axis acceleration limit, trajectory smoothing coefficient, speed limit value, allowable load level or task receiving priority identifier. For actions like load distribution that require coordination with the upper-level scheduler, the management terminal can send load limiting or load reduction parameters related to the local machine to the controller on the one hand, and synchronize scheduling suggestions to the production line scheduling module on the other hand. Among these, the control adjustment information sent to the controller is still directly related to the local machine's operating constraints, thus maintaining consistency with the system structure of this embodiment.

[0024] Furthermore, the aforementioned ideal dynamic reference data, theoretical abnormal response data, and operational feedback data are preferably checked for correspondence before participating in the differential analysis. That is, position feedback corresponds to position reference, current feedback corresponds to current reference, torque feedback corresponds to torque reference, and vibration feedback corresponds to vibration reference. Differentiation is only performed between semantically consistent response quantities of the same type to avoid misjudgment caused by mixing and comparing data with different physical meanings. The purpose of this step is to separate changes in operating conditions from failure degradation as much as possible through a closed-loop comparison link of control commands, ideal benchmarks, theoretical anomalies, and actual feedback, thereby achieving early failure prediction and health management under varying operating conditions.

[0025] In a preferred embodiment of the present invention, the monitoring server is further configured to: obtain an ideal dynamic model from a pre-configured memory or database; input control command data into the ideal dynamic model; and output ideal dynamic reference data based on the ideal dynamic model; wherein the ideal dynamic model is used to characterize the response relationship under frictionless, wearless, and transmission clearance-free conditions, and when the structural stiffness parameters are at reference values.

[0026] This embodiment provides an ideal dynamic reference reconstruction step; specifically, in the aforementioned long-term operation scenario of a five-axis CNC machining center, actual feedback data alone is insufficient to distinguish between normal heavy load and abnormal degradation, because there are significant inherent differences in the current and vibration amplitude of different machining sections. If the historical average value is used directly as a reference, the average reference will be significantly distorted when the tool path switches from circular interpolation to rapid acceleration and deceleration linear interpolation. Therefore, this embodiment introduces a mechanistic ideal dynamic model as an absolute reference. Specifically, the monitoring server reads the ideal dynamic model corresponding to the current model from a pre-configured memory or database. This model can be modeled separately for each axis and then output together. For example, a position-velocity-current response relationship can be established for the X-axis, a posture change-torque demand relationship can be established for the A / C axes, and a speed setting-torque response relationship can be established for the spindle. The model parameters here all use the factory calibration values ​​or the reference values ​​after overhaul. By default, the friction term is zero, the wear term is zero, the transmission clearance is zero, and the structural stiffness is the reference stiffness. Regarding data input, control command data is organized into a timing input sequence that matches the model. Assuming there are four sampling points, the X-axis position command is [0, 1, 2, 3] mm, the velocity command is [0, 50, 50, 0] mm / s, and the acceleration command can be simplified to [0, 1, -1, 0]. Then the model outputs a set of ideal reference sequences, such as the ideal position response [0, 1, 2, 3] and the ideal current response [0.0, 1.8, 1.8, 0.0]. If the A-axis attitude change is also considered, the model can also output ideal torque references [0.0, 0.6, 0.6, 0.1]; these reference values ​​constitute the ideal dynamic reference data within the current time window.

[0027] Furthermore, the model can adopt a hierarchical structure: the upper layer generates ideal kinematic references based on the trajectory and interpolation period, and the lower layer obtains ideal current, torque and displacement responses based on motor constants, inertia and stiffness parameters; in order to meet engineering feasibility, the ideal dynamic model does not need to be constructed as a complex continuous equation, and can also be implemented by combining table lookup and discrete state update; for example, the reference parameter sets corresponding to the rapid traverse mode, cutting mode and fine positioning mode of the same machine tool can be pre-stored for different working modes, and then switched according to the current mode of the controller; As a fault-tolerant implementation mechanism, if there is no ideal model in the database that matches the current model, shaft configuration, or installation fixture, the server can revert to the previous verified model and mark the model as compatible for operation; if the control command lacks an acceleration term, it can be estimated by the difference between adjacent velocity samples; if the command span of a certain sampling segment exceeds the preset span threshold and exceeds the model's stable range, the segment is split into multiple sub-segments before being input into the model to avoid excessive single interpolation span causing reference distortion. For example, during the semi-finishing stage of the impeller, the program segment switches from uniform speed cutting to corner deceleration, and the controller continuously outputs the joint trajectory command of the X-axis and C-axis; the server retrieves the ideal model corresponding to the machine tool from the database, inputs the multi-axis command sequence within the current forty milliseconds, and generates the ideal position following curve and the ideal servo current curve. At this point, although the actual current increases due to changes in the local hardness of the material, the ideal reference still changes synchronously with the trajectory, so that the subsequent residuals mainly reflect the deviations caused by structural degradation rather than changes in the process operation itself. The purpose of this step is to provide a dynamic reference that is strictly aligned with the current operating conditions for subsequent anomaly identification, thereby establishing a basis for comparable residual analysis under different trajectories and load conditions.

[0028] In a preferred embodiment of the present invention, the monitoring server is further configured to: retrieve a degradation mechanism library from a pre-configured memory or database; extract degradation mechanism parameters from the degradation mechanism library; input the degradation mechanism parameters into an ideal dynamic model, or modify the ideal dynamic model based on the degradation mechanism parameters; and generate theoretical abnormal response data based on the modified ideal dynamic model and control command data.

[0029] This embodiment provides a degradation mechanism injection step; specifically, an ideal benchmark alone cannot directly indicate which mechanism the fault belongs to, because the actual residual may simultaneously contain multiple factors such as wear, thermal drift and load fluctuation. If the actual residual is only compared with the fixed fault template, the template will be out of the current instruction context and it will be difficult to remain effective in complex trajectories. Therefore, this embodiment superimposes degradation mechanism parameters on the ideal dynamic model to actively deduce the theoretical abnormal response characteristics of a specific fault under the current trajectory. Specifically, the monitoring server retrieves multiple sets of degradation parameters from a degradation mechanism library. Each set is not an isolated label, but a parameterized description that can be injected into the model. For example, one set may represent X-axis ball screw wear, which may include a friction change parameter of 0.15 and a backlash change parameter of 0.03mm. Another set may represent C-axis drivetrain stiffness reduction, which may include a stiffness reduction ratio of 8%. Yet another set may represent servo motor demagnetization, which may include a torque constant reduction ratio of 5%. Regarding the injection method, two implementation paths can be adopted. The first is to directly use the degradation parameters as model input and enter the state update process together with the control command. The second is to first correct the internal coefficients of the model based on the degradation parameters, and then use the corrected model to calculate the abnormal response. Both methods can be implemented in engineering. Their common point is that the output is the theoretical abnormal response data under the same trajectory and the same control cycle. As a specific data processing example, suppose the ideal model outputs an ideal position [5.00, 5.40, 5.80] and an ideal current [1.2, 1.5, 1.7] on the X-axis at a certain time window; if the friction variation parameter is injected by +0.2, the model believes that a larger driving force is required under the same speed command, and the abnormal current can rise to [1.3, 1.8, 2.1]; if the gap variation parameter is further added by +0.04mm, the position response of the reverse reversing segment will produce a slight hysteresis, and the abnormal position can become [4.99, 5.37, 5.75]; this set of abnormal positions and abnormal currents constitutes the theoretical abnormal response of this mechanism at this time window; In actual deployment, the server can traverse the degradation mechanism library, or it can first filter candidate mechanisms by axis system, working condition and historical statistical results; for example, if the cumulative overload times of the X-axis in the current batch exceed the preset overload threshold or are higher than the preset proportion of other axes, the parameter groups related to the X-axis lead screw, guide rail and servo motor will be extracted first to reduce the amount of real-time inference. In abnormal situations, if multiple degradation mechanism parameters are applicable to the current time window, the server can first independently generate multiple sets of single-fault theoretical abnormal responses, and then generate some dual-fault combined responses depending on the computing resources. If a set of parameters in the mechanism library lacks a valid calibration range, it will not participate in online judgment and will only be retained as a candidate for offline analysis. If the injection of a parameter causes the model output to be unstable, such as non-physical jumps occurring in multiple consecutive frames, the parameter set will be automatically rolled back and recorded as a mechanism to be revised. For example, during the roughing stage of the impeller, the X-axis bears a large feed load, while the A-axis performs small-angle compensation simultaneously. The server prioritizes extracting three sets of mechanisms from the degradation mechanism library: X-axis lead screw wear, X-axis servo demagnetization, and A-axis transmission chain clearance. Each set is then injected into the current multi-axis ideal model. Finally, three corresponding theoretical abnormal response curves are obtained, which are used to compare with the actual residuals one by one. In this way, even if the roughing current itself is high, the deviation patterns that should exist for different mechanisms can be observed under the same trajectory, speed, and working conditions. The purpose of this step is to transform empirical fault knowledge into computable parameterized objects, thereby enabling active simulation and distinguishable modeling of different degradation mechanisms.

[0030] In a preferred embodiment of the present invention, the degradation mechanism parameters include at least one of friction variation parameters, clearance variation parameters, stiffness variation parameters, and torque constant variation parameters.

[0031] This embodiment provides a mechanism for refining the configuration of degradation mechanism parameters. Specifically, if the above-mentioned mechanism injection only uses macroscopic fault labels, there is still a deficiency in the ability to identify the faults. For example, the same X-axis anomaly may correspond to increased guide rail friction, increased lead screw backlash, decreased stiffness caused by loose mounting components, or degradation of the magnetic properties of the servo motor. Different mechanisms have different influence paths on current, position, and vibration. If the mechanism is not refined to the parameter level, the identifiability of the theoretical anomaly response will be significantly reduced. Therefore, this embodiment decomposes the degradation mechanism into at least one specific parameter. Specifically, the friction variation parameter is used to characterize the increase in transmission pair resistance, which can be a fixed increment or a piecewise parameter that varies with speed; the clearance variation parameter is used to characterize the dead zone expansion during commutation; the stiffness variation parameter is used to characterize the increase in following error caused by the softening of the force-displacement relationship; the torque constant variation parameter is used to characterize the decrease in output torque under the same current; assuming that in the same commutation action, the X-axis position error is [0.00, 0.01, 0.00] and the current is [1.0, 1.2, 1.0] under ideal conditions; if only the friction variation parameter increases, the position error may become [0.00, 0.02, 0.01] and the current may become [1.2, 1.5, 1.3], characterized by the overall current amplitude being greater than the preset current reference; If only the gap variation parameter increases, the position error may become [0.00, 0.05, 0.00], and the current fluctuation is less than the preset fluctuation threshold. Its characteristic is that the maximum position error exceeds the preset error threshold during commutation. If only the stiffness variation parameter decreases, the position error may become [0.01, 0.03, 0.02], and the vibration feedback is amplified. Its characteristic is that the continuous deviation increases under load. If only the torque constant variation parameter decreases, the current will also rise, but it will be more obvious in the heavy load section, and the proportional relationship between torque feedback and current will change. The server can construct a more discriminative theoretical abnormal response based on this.

[0032] Furthermore, these parameters can be used individually or in combination; for example, long-term wear of the lead screw often leads to changes in both friction and clearance; bearing aging may manifest as increased friction or a decrease in local stiffness; to avoid an excessive number of combinations, the system can preset priorities, such as screening single parameters first, and then performing combined deduction on two sets of parameters with similarity. As a fault-tolerant implementation mechanism, if a parameter is not sensitive in the current sampling window, such as in a pure uniform speed without commutation window where the gap change parameter is difficult to fully manifest, the server can reduce the judgment weight of that parameter group instead of directly giving the conclusion that it does not exist; if multiple parameter combinations lead to approximately the same theoretical residual, they are retained as the same candidate mechanism group, and will be further judged in a more discernible working condition window later. For example, in one step of impeller finishing, the X-axis performs a small reciprocating correction; the server finds that the position residual in this time window has a peak at the reversal point, while the current increase is not large, so it prioritizes the matching priority of the clearance change parameter group and weakens the effect of the friction change parameter; in the subsequent heavy-load cutting time window, the current is observed to rise continuously, so the friction change parameter is superimposed, and finally a more refined diagnostic result is formed that the wear of the lead screw causes the clearance and friction to increase together; The purpose of this step is to enhance the separability of various degradation mechanisms through parametric modeling, thereby achieving more accurate fault location and stage judgment.

[0033] In a preferred embodiment of the present invention, the monitoring server is further configured to: generate actual residual data based on operational feedback data and ideal dynamics benchmark data; generate theoretical residual data based on theoretical abnormal response data and ideal dynamics benchmark data; and use the actual residual data and theoretical residual data as input data for similarity verification.

[0034] This embodiment provides a dual-track residual construction step. Specifically, in the aforementioned scheme, if the actual operation feedback is directly compared with the theoretical abnormal response, there will be a significant drawback: both contain a large number of common components related to the current trajectory itself, such as the current in the acceleration segment inevitably increasing and the position curve in the commutation segment inevitably bending. Direct comparison can easily lead to mistaking normal motion behavior for fault characteristics. Therefore, this embodiment first subtracts the common ideal dynamic reference data separately, and then performs similarity verification. Specifically, the actual residual data is obtained by subtracting the ideal dynamic reference data from the operational feedback data, and the theoretical residual data is obtained by subtracting the ideal dynamic reference data from the theoretical abnormal response data. Both share the same reference, which allows subsequent comparisons to focus on the parts that deviate from the ideal state. In order to make the residuals uniform and comparable, the dimensions of each channel can be normalized first. For example, the position error can be normalized according to the allowable following error, the current deviation can be normalized according to the rated current, and the vibration deviation can be normalized according to the sensor range. Suppose the original actual residuals of the three channels within a certain time window are position [-0.02, -0.04] mm, current [0.2, 0.5] A, and vibration [0.01, 0.03] g, respectively. If the corresponding normalized scales are 0.10 mm, 2.0 A, and 0.10 g, then the normalized actual residuals can be written as [-0.2, -0.4], [0.1, 0.25], and [0.1, 0.3]. The theoretical residuals are processed similarly to facilitate unified input to the similarity module. In terms of data organization, the multi-channel residuals of each time window can be stacked into a two-dimensional matrix. Taking the above two sampling points and three channels as an example, the actual residual matrix can be written as two rows and three columns. The first row corresponds to the three normalized deviations of t1, and the second row corresponds to the three normalized deviations of t2. The theoretical residual matrix also adopts the same structure. Subsequently, whether time warping or vector similarity algorithm is used, these two matrices are used as input.

[0035] Furthermore, to ensure the stability of the physical meaning of the residuals, the residuals are preferably constructed according to the principles of coaxiality, same response quantity, and same time window; that is, the X-axis position feedback is only subtracted from the X-axis ideal position reference, and the X-axis current feedback is only subtracted from the X-axis ideal current reference. If the theoretical anomaly response includes A-axis, C-axis, or principal axis channels, these channels are used to independently construct theoretical residuals with their corresponding ideal references. Multi-axis, multi-channel residuals can be concatenated in a predetermined order before entering the similarity module, such as first position-type channels, then current-type channels, and then vibration-type channels, to ensure consistent input structures across different windows. If a certain type of feedback quantity is not suitable for direct point-by-point subtraction with the instantaneous reference, such as when high-frequency vibration waveforms are more commonly represented by statistical features like envelope values, root mean square values, or frequency band energy in engineering, vibration feature quantities corresponding to the ideal model output can be extracted within the current time window and then subtracted from the vibration reference features to form residuals. This approach does not alter the basic chain of feedback data subtracting reference data and avoids noise amplification issues caused by direct difference of the original high-frequency waveforms. As a fault-tolerant implementation mechanism, if a sensor in a certain channel temporarily fails, the residual of that channel is marked as empty and is calculated by the similarity module using a subset of effective channels; if the number of effective channels is lower than the minimum requirement, for example, if only one low-confidence channel remains, the determination of this time window is suspended and a resampling is requested; if the normalization scale is zero or abnormally close to zero, a preset minimum scale is used instead to avoid residual distortion caused by division by zero. For example, when performing a corner transition in a machining center, the actual current increase and vibration enhancement are normal phenomena. The server first subtracts the ideal reference from these actual responses to obtain the actual residuals that only reflect deviations from the ideal behavior. At the same time, the abnormal response under the assumption of wear on the X-axis is also subtracted from the same ideal reference to obtain the theoretical residuals. In this way, subsequent comparisons are no longer affected by the corner action itself, but focus on whether there is an additional deviation consistent with the wear mechanism. The purpose of this step is to eliminate the influence of common patterns caused by actual process operations, thereby achieving purified expression and stable comparison of degenerative features.

[0036] In a preferred embodiment of the present invention, the monitoring server is further configured to: calculate the temporal similarity between the actual residual data and the theoretical residual data after time alignment based on a dynamic time warping algorithm, or calculate the similarity between the residual feature vectors of the same dimension corresponding to the actual residual data and the theoretical residual data based on a multidimensional spatial cosine similarity algorithm, to obtain a similarity; determine that the degradation mechanism corresponding to the theoretical residual data exists in response to the similarity being greater than or equal to a preset threshold; determine that the degradation mechanism corresponding to the theoretical residual data does not exist in response to the similarity being less than the preset threshold; wherein, the preset threshold is obtained by offline calibration based on the similarity distribution of historical normal samples and historical fault samples.

[0037] This embodiment provides a residual similarity verification mechanism. Specifically, although the ideal working condition component has been removed in the previous step, there may still be slight misalignment between the actual residual and the theoretical residual in time. For example, there may be a millisecond delay between encoder sampling and vibration sampling, or the peak time may be slightly shifted due to a decrease in stiffness of the actual equipment. If point-by-point absolute time alignment comparison is still used, the true similarity is easily underestimated. Therefore, this embodiment introduces time alignment and multi-dimensional similarity strategies before similarity calculation. When focusing on the time evolution of the residual waveform, a dynamic time warping algorithm can be used. For ease of explanation, assume the actual current residual sequence is [0.1, 0.4, 0.6, 0.3] and the theoretical current residual sequence is [0.1, 0.3, 0.6, 0.4, 0.2]. Although the theoretical sequence has one more sampling point, the peak evolution trends of the two are similar. After time alignment, a minimum cost matching path can be obtained, and the warping distance is output. This distance is then converted into a similarity score, for example, by subtracting the normalized distance from 1 to obtain a similarity score of 0.86. If this value is higher than the threshold of 0.80, it can be considered that the mechanism has a strong match on this channel. When the multi-channel residuals have achieved frequency synchronization and more attention is paid to the overall directional consistency, a multi-dimensional spatial cosine similarity algorithm can be used. Assuming that a certain time window extracts three residual features of position, current, and vibration through the root mean square value calculation, forming an actual feature vector [0.3, 0.4, 0.5] and a theoretical feature vector [0.2, 0.5, 0.45], the angle between the two is small, and the cosine value can be close to 1. If the calculated value is 0.93 and exceeds the preset threshold of 0.85, then the degradation mechanism is determined to exist.

[0038] The threshold is not set subjectively on a temporary basis, but is obtained through offline calibration based on historical normal samples and historical fault samples. Specifically, a large number of non-fault residual similarity distributions can be collected during the normal operation of the machine tool, and the normal range is mostly below 0.55. Then, the similarity distribution of known faults can be collected from the test bench or historical maintenance records, and the fault range is mostly above 0.78. The system can then set the range between 0.70 and 0.80 as the candidate threshold, and then determine the final threshold by calculating the intersection point corresponding to the minimum sum of the false alarm rate and the false alarm rate, such as 0.76. This can take into account both early identification capability and false alarm control.

[0039] Furthermore, to avoid misjudgment due to similarity on a single channel but dissimilarity overall, the preferred similarity in this embodiment is the result of multi-axis and multi-channel integration within the current time window. Channel similarity can be calculated separately for each effective channel, such as position, current, torque, and vibration, and then fused according to preset weights to obtain the total similarity for threshold comparison. The weights can be preset based on channel credibility, the sensitivity of the mechanism to that channel, or the identifiability of the current operating condition. For example, for the lead screw wear mechanism, the weights of the position and current channels can be increased, while the weight of the vibration channel, which has lower identifiability in the current window, can be decreased. Thus, the aforementioned single-channel example is only used to illustrate a certain calculation process, and the final existence determination is still based on the comprehensive similarity.

[0040] Furthermore, the aforementioned time alignment optimization includes two layers of meaning: one is resampling or timestamp correction of different acquisition sources on a unified time base, and the other is local elastic alignment after entering the dynamic time warping algorithm; the former is used to eliminate fixed or slowly changing delays caused by the acquisition link, and the latter is used to absorb the slight forward or backward shift of peaks caused by equipment degradation; after such processing, the dynamic time warping and cosine similarity algorithms correspond to two types of application scenarios that focus on temporal pattern and multi-dimensional direction, respectively, avoiding unclear algorithm selection logic; In abnormal situations, if the highest similarity falls into the pending judgment range, such as between 0.72 and 0.76, the server will not directly give a conclusion of existence or non-existence, but will mark the current mechanism as pending confirmation and require the accumulation of the next N time windows before voting; where N is the preset number of consecutive discrimination time windows, used to represent the number of time windows to continue to accumulate after the current pending confirmation result, for example, 3, 5 or 7; If the similarity of multiple mechanisms exceeds the threshold, they are sorted by similarity and filtered based on mutual exclusion relationships. If the similarity of all mechanisms is below the threshold, they are identified as outside the current known mechanism library, and the mechanism is output as unmatched while the residual is retained for subsequent mechanism library updates. For example, in ten consecutive tool position windows of impeller finishing, the server calculates the similarity between the actual residual and three types of theoretical residuals: lead screw wear, spindle bearing abnormality, and servo demagnetization. The results show that the similarity of lead screw wear in the seven windows is 0.81, 0.84, 0.79, 0.83, 0.80, 0.82, and 0.85, respectively, while the other two types are always below 0.65. Even if individual windows fluctuate due to sudden cutting changes, as long as most windows continuously exceed the threshold, the system can still stably identify it as this type of degradation mechanism. The purpose of this mechanism is to reduce the impact of sampling misalignment and amplitude disturbance by judging the consistency of temporal morphology and multidimensional direction, thereby achieving reliable verification of weak degradation features.

[0041] In a preferred embodiment of the present invention, the monitoring server is further configured to: determine the fault type by matching the similarity with a preset fault type feature library when it is determined that the degradation mechanism corresponding to the theoretical residual data exists; determine the fault stage based on the trend mapping of the degradation mechanism parameters; and calculate the remaining lifetime assessment result by extrapolating the data using the pre-acquired historical degradation trajectory; and determine the current state as a non-fault fluctuation state when it is determined that the degradation mechanism corresponding to the theoretical residual data does not exist; wherein, the health status result includes at least one of the fault type, fault stage, remaining lifetime assessment result, and non-fault fluctuation state.

[0042] This embodiment provides a health status result generation mechanism. Specifically, simply outputting the existence of a certain mechanism is insufficient to support workshop maintenance decisions. It is necessary to know the fault type, the current degradation stage, and the remaining service life. If these results are missing, although the system has monitoring capabilities, it cannot form closed-loop management. Therefore, after determining the existence of the mechanism, this embodiment continues to generate fault type, fault stage, and remaining service life assessment results. If the mechanism does not exist, it is clearly marked as a non-fault fluctuation state. Specifically, the fault type can be obtained by matching the similarity results with the fault type feature library; this feature library can be organized into three levels: shaft system—component—mechanism; for example, X-axis—lead screw pair—wear, C-axis—reducer—increased clearance, spindle—bearing—decreased stiffness, etc.; the server compares the current optimal matching mechanism and its multi-window similarity distribution with the feature library and outputs the specific fault type; it should be further explained here that matching the similarity with the preset fault type feature library is not just taking a single similarity scalar in isolation to search for tags, but using the currently identified degradation mechanism as the main index, and combining the similarity results of the mechanism on multiple time windows, multiple channels or multiple shaft systems for matching; The fault type feature library pre-stores similarity rules, channel sensitivity order, shaft system affiliation, and mechanism mapping relationship corresponding to each fault type. The server uses the mechanism identifier and similarity result as joint input for matching, thereby stably mapping the existence result of the degradation mechanism to the fault type. After such processing, the aforementioned description of the similarity matching fault type feature library is consistent with the specific implementation in this embodiment, avoiding the misunderstanding of similarity as a single value detached from the mechanism context. The failure stage can be mapped based on the changing trends of degradation mechanism parameters. Assuming the system continuously monitors the X-axis friction parameter increasing from 0.05 to 0.10 and then to 0.18 within a week, while the clearance parameter increases from 0.00 to 0.02 and then to 0.04, then 0 to 0.08 can be defined as the early stage, 0.08 to 0.15 as the development stage, and greater than 0.15 as the aggravation stage. Thus, the current stage can be mapped to the late development stage. The remaining life assessment result can be extrapolated using historical degradation trajectories. Assuming that historically, similar machine tools typically require replacement when the lead screw friction parameter reaches 0.30, and it has currently reached 0.18, with the parameter sequence for the last five days being [0.10, 0.12, 0.14, 0.16, 0.18], the average daily increase is approximately 0.02. Linear extrapolation estimates approximately 6 days remaining until the maintenance threshold is reached. In practical applications, more robust life estimation methods such as piecewise slope, exponential fitting, or similar trajectory matching can also be used.

[0043] When the similarity of all known mechanisms is below the threshold, or when there are fluctuations but no consistent mechanism correspondence, the server determines the current state as a non-fault fluctuation state. This state does not mean that it is completely normal, but rather that the existing evidence is closer to non-fault factors such as load changes, environmental disturbances, and material inhomogeneity. At this time, the system can suggest continued observation instead of immediate repair. As a fault-tolerant implementation mechanism, if the fault type has been determined but the parameters required for fault stage mapping are insufficient, such as only friction trend and lack of gap trend, then partial results will be output first; if the number of historical trajectories used for lifetime extrapolation is insufficient, then an interval lifetime estimate will be given instead of a single value result, such as a remaining lifetime of 5 to 8 days; if the current state repeatedly switches between multiple fault types, then stability constraints will be enabled, and the final type will only be output when the same type exceeds a preset number of consecutive times. For example, in the third week of impeller batch processing, the server continuously identified the existence of the X-axis lead screw related mechanism and observed that the friction change parameters increased day by day; combined with the fault type feature library, the system determined it to be an X-axis lead screw pair wear type fault; combined with trend mapping, it was judged to be in the development stage; and then extrapolated using the similar degradation trajectory before each maintenance, the remaining life was obtained to be about 72 hours; if another period of time caused a short-term increase in current due to a local high hardness area of ​​the blank, but no stable matching mechanism, then this period of time was only marked as a non-fault fluctuation and did not trigger erroneous maintenance; The purpose of this mechanism is to transform similarity judgments into actionable equipment health status results, thereby expanding the functionality from identifying anomalies to supporting maintenance decisions.

[0044] In a preferred embodiment of the present invention, the management terminal is further configured to: output warning information when the fault stage is greater than or equal to the warning level set according to the fault stage division rules, and not output warning information when the fault stage is less than the warning level; output maintenance prompt information when the remaining life assessment result is less than or equal to the life threshold set according to the maintenance strategy, and output normal life prompt information when the remaining life assessment result is greater than the life threshold; and output continue running information when the current state is a non-fault fluctuation state.

[0045] This embodiment provides a hierarchical output mechanism. Specifically, if all identification results are uniformly displayed as alarms, a new drawback will arise: maintenance personnel may easily overlook truly high-risk events due to an excessive number of alarms. Therefore, this embodiment adopts different information output strategies based on the fault stage, remaining lifespan, and non-fault fluctuation state. Specifically, the management terminal has pre-stored fault stage division rules and maintenance strategies; the fault stage division rules can divide the degradation process into four levels: early stage, development, aggravation and critical stage, and set development or aggravation as the warning level; the maintenance strategy can set the life threshold, such as 48h, according to the production line cycle, spare parts delivery cycle and machine tool importance. When the stage result is greater than or equal to the warning level, the terminal outputs a warning message; when it is lower than the level, only the status change is recorded and no warning pops up; assuming the current stage codes are 1, 2, 3, and 4, corresponding to early stage, development stage, aggravation stage, and critical stage, the warning level is set to 2; if the current stage is 1, the panel only displays a yellow status label; if the current stage is 2 or higher, a warning window pops up and is pushed to the maintenance team. When the remaining lifespan is less than or equal to the lifespan threshold, the terminal outputs a maintenance prompt message; if the lifespan assessment is 36 hours and the threshold is 48 hours, it suggests checking and replacing the X-axis lead screw lubrication assembly during the next shift's downtime window; if the lifespan assessment is 120 hours, it outputs that the lifespan is normal and maintains the existing inspection strategy. When the status is determined to be a non-fault fluctuation, the terminal outputs a continue operation message; such messages can be explicitly stated as the current fluctuation is inconsistent with the known fault mechanism, and it is recommended to continue operation and keep an eye on it, so as to avoid the operator mistaking material fluctuations for equipment damage; As a fault-tolerant implementation mechanism, if there is a conflict between the stage results and the lifetime results, for example, if the stage is still in the early stage but the lifetime estimate is already below the threshold, a more conservative strategy will be adopted to output maintenance prompts; if the lifetime estimate is unavailable, warnings will be output only based on the stage; if the terminal and server communication is interrupted for a short time, the terminal will cache the most recent valid result and resend the event records that were missed during the connection restoration period. For example, when the impeller finishing batch is nearing delivery, the system determines that the X-axis lead screw wear has entered the development stage, with a remaining lifespan of approximately 40 hours. Since the warning level is set to the development stage and the lifespan threshold is 48 hours, the terminal immediately pushes a warning message to the team leader and equipment engineer, and at the same time generates a maintenance prompt to check the lead screw lubrication and confirm the spare parts inventory during the tool change window tonight. Conversely, in another current fluctuation caused only by cutting a local high-hardness area, the terminal only displays the information to continue running, without interrupting the current work order cycle. The purpose of this mechanism is to differentiate the display and response methods based on the level of risk, thereby achieving less disruption and more relevant operation and maintenance interactions.

[0046] In a preferred embodiment of the present invention, the management terminal is further configured to: generate control adjustment information based on health status results; send control adjustment information to the controller; wherein the control adjustment information is used to adjust the operating speed, acceleration or load distribution, and the maintenance prompt information is used to prompt adjustments to the maintenance cycle or maintenance plan.

[0047] This embodiment provides a control adjustment closed-loop mechanism; specifically, simply providing alarms and maintenance suggestions still has a limitation: when spare parts are not yet available or the current batch of tasks cannot be stopped immediately, the equipment needs to be operated under control in a state with early degradation characteristics; if there is no flexible adjustment at the control level at this time, the degradation may deteriorate rapidly; therefore, this embodiment automatically generates control adjustment information based on the health status results and sends it to the controller to form a closed loop of monitoring-judgment-regulation. Specifically, the management terminal can have a built-in control strategy table to map different health status results to different control actions; for example, when the X-axis lead screw is detected to have entered the development stage, the terminal issues a command to reduce the maximum X-axis acceleration from 1.0g to 0.7g; when the spindle bearing stiffness is detected to have decreased, the terminal issues a command to limit the maximum spindle speed to 85% of the rated value. When the lifespan of a certain axis approaches the threshold, load limiting parameters, load reduction parameters, or task receiving priority identifiers directly related to the machine can be generated, and these parameters can be sent to the controller as control adjustment information to reduce the current equipment load. Assuming that the health status result includes fault type = X-axis screw wear, stage = development, and remaining lifespan = 36h, the control adjustment information can be generated as a triplet: running speed adjustment coefficient 0.9, acceleration adjustment coefficient 0.7, and load limit level 0.6. After receiving the data, the controller will correct the relevant parameters of the X-axis according to the above coefficients in the subsequent program segment, or refuse to accept high-load processing segments that exceed the corresponding load limit level according to the load limit level. Here, the load allocation is preferably understood as the load constraints, load limit levels or task acceptance priorities that the controller can directly execute, rather than directly equating the external scheduling actions that require production line-level rearrangement with the control adjustment information sent to the controller.

[0048] Furthermore, if the workshop also has a higher-level scheduling system, the management terminal can send the above-mentioned control adjustment information to the controller, and at the same time, send the suggestion to relocate the heavy cutting process to other equipment as a maintenance scheduling suggestion to the production line scheduling module. However, the maintenance scheduling suggestion and the control adjustment information sent to the controller are distinguished from each other in terms of object and execution level. The information sent to the controller is still directly executable parameters or limits such as speed, acceleration, rotation speed limit, allowable load level, and task receiving priority identifier, so as to maintain consistency with the system structure of this embodiment. Maintenance reminders can also be updated synchronously; for example, if the original maintenance cycle was once every two weeks, and the current health status results indicate that the degradation has entered the development stage, the maintenance cycle can be dynamically adjusted to check the lubrication status once per shift, review the residual trend once every 24 hours, and insert the spare parts procurement plan in advance. As a fault-tolerant implementation mechanism, if the controller refuses to receive online adjustment parameters, the terminal will convert the control adjustment information into a manual confirmation instruction, which will be manually activated by the operator at the safe tick point; if multiple adjustment strategies conflict with each other, such as requiring both a reduction in speed and an increase in tick rate, the load reduction strategy can be retained according to the principle of safety priority; if the similarity continues to rise rapidly after adjustment, the terminal will upgrade its status and recommend unplanned shutdown. For example, during a weekend night shift of impeller mass production, the system determines that the X-axis lead screw wear has entered a development stage, but there are still two batches that must be completed. On the one hand, the terminal sends speed and acceleration reduction commands to the controller to reduce commutation impact; on the other hand, it can synchronize maintenance scheduling suggestions to reduce the proportion of roughing tasks on the machine to the scheduling system. The machine controller actually receives the load limit and load reduction parameters. At the same time, the maintenance plan is changed from weekly inspection to next-day shutdown inspection. In this way, the equipment can complete emergency processing tasks under controlled risks and avoid the sudden expansion of the fault. The purpose of this mechanism is to directly translate diagnostic results into equipment behavior constraints and maintenance plan adjustments, thereby achieving proactive intervention in health management.

[0049] In a preferred embodiment of the present invention, the control command data includes at least one of PLC control logic data, multi-axis position commands, speed commands, and trajectory commands, and the operation feedback data includes at least one of servo current, encoder position, torque feedback value, and vibration feedback value corresponding to the output response of the ideal dynamic model.

[0050] This embodiment provides a data type configuration method; specifically, in order to enable the above scheme to adapt to different levels of industrial equipment, the system does not limit itself to using data from a single source, but allows at least one of the control side and feedback side to be selected to form an effective monitoring link; this setting can avoid the inability to deploy when the sensor configuration of some older equipment is insufficient, and is also compatible with multi-source fusion monitoring of high-end equipment; Specifically, the control command data may include at least one of PLC control logic data, multi-axis position commands, speed commands, and trajectory commands; for high-end five-axis machining centers, multi-axis position commands, speed commands, and trajectory commands are usually collected first; for automated loading and unloading units, PLC logic sequences, such as clamping, releasing, homing, and transport status switching signals, can also be collected. The operational feedback data is selected from the data corresponding to the output response of the ideal model. For example, servo current corresponds to the driving force requirement, encoder position corresponds to the actual following behavior, torque feedback value corresponds to mechanical load, and vibration feedback value corresponds to the structural dynamic state. Specifically, before the operational feedback data enters the residual construction link, a one-to-one correspondence should be established with the output response of the same name or physical meaning in the ideal dynamic model. For example, when a multi-axis position command outputs an ideal position reference through the model, the encoder position can be used as the operational feedback. When the speed command is derived from the model to produce the ideal current or ideal torque response, the corresponding servo current or torque feedback value can be used; the trajectory command or PLC control logic data is used to describe the current action stage and working context, and it is mainly used as model input, and is not directly subtracted from the encoder position, servo current, torque feedback value or vibration feedback value. The role of PLC control logic data and trajectory instructions is to constrain the model to output what ideal response under what operating conditions, while residual calculation still occurs between the running feedback data and the output response of the ideal dynamic model, thus maintaining consistency with the residual construction logic in the aforementioned embodiments. If the current equipment can only stably provide X-axis position commands, speed commands, servo current, and encoder position, then the ideal model can at least output ideal position and ideal current. The system can still generate position residuals and current residuals and complete health judgments. If vibration sensors are added later, the vibration channels can be expanded on the original basis, improving the ability to identify stiffness reduction-type faults. For example, for a turntable station with PLC cycle control, the logic sequence of clamping completion—rotation—release completion can be used as part of the trajectory context, so that the model knows when a load change should occur, thereby reducing misjudgments.

[0051] In terms of data alignment, the system can uniformly resample data with different sampling frequencies to the same time grid; for example, if position and current are sampled at 1ms and vibration is sampled at 0.2ms, the vibration can be extracted by windowing feature values ​​and aligned to the 1ms grid; if a certain type of control instruction updates at a low frequency, such as PLC logic updating once every 10ms, the state remains unchanged within its effective range for reference by the model and residual module. As a fault-tolerant implementation mechanism, if the control command data and the operational feedback data cannot correspond semantically, for example, if the model only outputs the position reference while the acquisition end only provides the ambient temperature, then the data will not enter the current decision link; if the feedback quantity has abnormal saturation or jamming, for example, the encoder position remains unchanged and the current is always at full scale, then it will first be judged as an acquisition failure and the channel will be isolated; if the available channels are reduced to below the minimum configuration, the system will prompt that the data conditions are insufficient, only provide trend observation, and not output a formal health conclusion. For example, in the impeller machining scenario throughout the text, the server mainly uses multi-axis position commands, speed commands, and trajectory commands as control inputs, and mainly uses servo current, encoder position, and vibration feedback as operational feedback. When the vibration acquisition card is temporarily offline during the night shift, the system still uses the position and current channels to maintain monitoring. After the vibration channel is restored, it is then included in the similarity calculation again. In this way, continuous health management capabilities can be maintained even in a production environment without stopping the machine. The purpose of this step is to adapt the solution to different device conditions and sensor configurations through flexible data type combinations, thereby enabling the solution to be deployed and operated stably.

[0052] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent equipment monitoring system with fault prediction and health management functions, characterized in that, include: Intelligent equipment, instruction data acquisition terminal, operation feedback acquisition terminal, monitoring server and management terminal; The intelligent equipment includes a controller, and an instruction data acquisition terminal is connected to the controller to acquire control instruction data and send it to the monitoring server. The operation feedback acquisition terminal is used to acquire operation feedback data of intelligent equipment and send it to the monitoring server; The monitoring server is used to generate ideal dynamics baseline data based on a preset ideal dynamics model and control command data, extract degradation mechanism parameters from a preset degradation mechanism library and generate theoretical abnormal response data by combining the ideal dynamics model and control command data, generate actual residual data based on the operation feedback data and ideal dynamics baseline data under the same time sequence, generate theoretical residual data based on the theoretical abnormal response data and ideal dynamics baseline data, and determine the health status result based on the similarity verification between the actual residual data and the theoretical residual data. The management terminal is used to receive and display health status results, output maintenance prompts, and send control adjustment information to the controller of the intelligent equipment.

2. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 1, characterized in that, The monitoring server is also used for: Retrieve a pre-defined ideal dynamic model from memory or database; Input the control command data into the ideal dynamic model; Output ideal dynamic benchmark data based on the ideal dynamic model; Among them, the ideal dynamic model is used to characterize the response relationship under the conditions of no friction, no wear, no transmission backlash, and structural stiffness parameters at the reference value.

3. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 1, characterized in that, The monitoring server is also used for: Retrieve a pre-defined degradation mechanism library from memory or database; Extract degradation mechanism parameters from the degradation mechanism library; Input the degradation mechanism parameters into the ideal dynamic model, or modify the ideal dynamic model based on the degradation mechanism parameters; Theoretical anomaly response data is generated based on the revised model and control command data.

4. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 3, characterized in that, The degradation mechanism parameters include at least one of the following: friction variation parameters, clearance variation parameters, stiffness variation parameters, and torque constant variation parameters.

5. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 1, characterized in that, The monitoring server is also used for: Actual residual data is generated based on operational feedback data and ideal dynamic baseline data; Theoretical residual data is generated based on theoretical anomaly response data and ideal dynamic benchmark data; The actual residual data and the theoretical residual data are used as input data for similarity verification.

6. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 1, characterized in that, The monitoring server is also used for: The time-series similarity between the actual residual data and the theoretical residual data after time alignment is calculated based on the dynamic time warping algorithm, or the similarity between residual feature vectors of the same dimension is calculated based on the multidimensional spatial cosine similarity algorithm. In response to a similarity greater than or equal to a preset threshold, the existence of a degradation mechanism corresponding to the theoretical residual data is determined; If the similarity is less than a preset threshold, it is determined that the degradation mechanism corresponding to the theoretical residual data does not exist; The preset threshold is obtained by offline calibration based on the similarity distribution between historical normal samples and faulty samples.

7. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 6, characterized in that, The monitoring server is also used for: Given that the degradation mechanism corresponding to the theoretical residual data exists, the fault type, fault stage, and remaining life assessment results are determined based on the similarity results, the changing trend of the degradation mechanism parameters, and the pre-acquired historical degradation trajectory. If it is determined that the degradation mechanism corresponding to the theoretical residual data does not exist, the current state is determined to be a non-fault fluctuation state; The health status results include at least one of the following: fault type, fault stage, remaining life assessment results, and non-fault fluctuation status.

8. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 7, characterized in that, The management terminal is also used for: When the fault stage is greater than or equal to the warning level set according to the fault stage division rules, a warning message is output; when the fault stage is less than the warning level, no warning message is output. When the remaining life assessment result is less than or equal to the life threshold set according to the maintenance strategy, a maintenance prompt message is output; when the current state is a non-fault fluctuation state, a continue operation message is output.

9. The intelligent equipment monitoring system with fault prediction and health management functions according to claim 1, characterized in that, The management terminal is also used for: Generate control adjustment information based on health status results; Send control adjustment information to the controller of the intelligent equipment; Among them, the control adjustment information is used to adjust the operating speed, acceleration or load distribution, and the maintenance prompt information is used to prompt adjustments to the maintenance cycle or maintenance plan.

10. The intelligent equipment monitoring system with fault prediction and health management functions according to any one of claims 1-9, characterized in that, The control command data includes at least one of PLC control logic data, multi-axis position commands, speed commands, and trajectory commands. The operation feedback data includes at least one of servo current, encoder position, torque feedback value, and vibration feedback value corresponding to the output response of the ideal dynamic model.

Images