An embedded monitoring method for tool wear of a numerical control machine tool

Abstract

The present application relates to the technical field of numerical control machine tool state monitoring, and particularly relates to an embedded monitoring method for tool wear of a numerical control machine tool. The method comprises the following steps: embedded monitoring of a tool in a machining process, synchronous multi-modal data acquisition and preprocessing, forming a real-time working condition data frame queue; cutting effective interval data interception on the real-time working condition data frame queue, obtaining effective cutting data sequences; discrete processing of current and tool feed rate on the effective cutting data sequences, forming an energy-efficiency benchmark fingerprint; extracting an expected healthy feed rate from the energy-efficiency benchmark fingerprint; performance deviation absolute amount calculation according to the expected healthy feed rate, and cutting performance attenuation analysis, obtaining a cutting performance attenuation degree. The present application realizes a complete process from detection to parameter self-adaptive closed-loop control on a resource-limited embedded system, prolongs tool life, guarantees machining quality, and improves production efficiency.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool condition monitoring technology, and in particular to an embedded monitoring method for CNC machine tool tool wear. Background Technology

[0002] Traditional tool monitoring methods using fixed thresholds suffer from severe low sensitivity. These methods typically monitor absolute thresholds for current or vibration, triggering an alarm only when the parameter exceeds a preset value. Since tool wear is a gradual process, this monitoring mechanism is almost undetectable for slow performance degradation, only triggering an alarm when the tool is severely worn or even chipped. By then, the workpiece is scrapped, resulting in material waste and production interruption. More seriously, the "normal value" varies greatly depending on the material and processing conditions, creating a dilemma in setting fixed thresholds: setting them too high leads to significant missed alarms, while setting them too low results in frequent false alarms.

[0003] Implementing advanced intelligent monitoring algorithms in resource-constrained embedded environments presents significant challenges. Advanced monitoring methods, such as machine learning and deep neural networks, require powerful computing capabilities and substantial memory resources. However, embedded controllers in industrial settings typically employ low-power microcontrollers with only a few hundred KB of RAM and limited Flash storage, making it difficult to support complex algorithms. Existing solutions either upload data to the cloud for analysis, leading to bandwidth pressure, poor real-time performance, reliance on network connectivity, and data security concerns; or drastically simplify the algorithm, sacrificing monitoring accuracy and reliability, creating a classic dilemma between "cloud intelligence and edge intelligence."

[0004] Multimodal data fusion and closed-loop control face significant technical barriers. Existing monitoring systems mostly perform single-parameter monitoring (e.g., monitoring only current or detecting only vibration), failing to comprehensively capture the complex wear characteristics of cutting tools. Furthermore, most are open-loop passive monitoring systems, only capable of issuing alarms and unable to actively adjust parameters, requiring manual operator intervention to respond to anomalies. In addition, the lack of cross-validation mechanisms between parameters leads to severe false alarms and missed alarms, significantly reducing the reliability and practicality of the monitoring system. Most critically, closed-loop control from anomaly detection to parameter adjustment requires complex decision-making logic, which is extremely difficult to implement in resource-constrained embedded environments.

[0005] In summary, existing technologies suffer from problems such as insufficient monitoring sensitivity, significant contradictions between algorithm complexity and embedded resources, and a lack of reliable multimodal fusion and closed-loop control capabilities, which urgently need to be addressed. Summary of the Invention

[0006] Therefore, it is necessary to provide an embedded monitoring method for CNC machine tool tool wear in order to solve at least one of the above-mentioned technical problems.

[0007] To achieve the above objectives, an embedded monitoring method for CNC machine tool tool wear includes the following steps:

[0008] Step S1: Embedded monitoring of the tool during the machining process, synchronous multimodal data acquisition and preprocessing to form a real-time working condition data frame queue; data of the effective cutting range is extracted from the real-time working condition data frame queue to obtain the effective cutting data sequence; the effective cutting data sequence is discretized for current and tool feed rate to form an energy-efficiency reference fingerprint.

[0009] Step S2: Extract the expected healthy feed rate from the energy efficiency benchmark fingerprint; calculate the absolute amount of performance deviation based on the expected healthy feed rate, and perform cutting performance degradation analysis to obtain the degree of cutting performance degradation;

[0010] Step S3: Calculate the vibration morphological anomaly degree of the real-time working condition data frame queue to obtain the vibration morphological anomaly degree; perform time-series analysis of tool wear trend based on cutting performance attenuation degree to obtain wear dynamic trend value; determine tool wear based on wear dynamic trend value and vibration morphological anomaly degree, optimize machining quality, and generate optimization strategy type; calculate feed rate adjustment according to optimization strategy type to obtain feed rate dynamic adjustment coefficient.

[0011] Step S4: Write the dynamic adjustment coefficient of the feed rate into the CNC controller of the CNC machine tool through the preset communication interface for closed-loop adjustment of machining parameters, and generate closed-loop adjustment instructions for machining parameters.

[0012] This invention systematically solves many shortcomings of existing technologies through a series of tightly coupled innovative methods. First, it overcomes the low sensitivity of traditional fixed-threshold monitoring. By learning and constructing a personalized energy-efficiency benchmark fingerprint for each new core component in step S1, and calculating the "cutting performance decay" in real time in step S2, it can capture early performance degradation caused by gradual component wear with extremely high sensitivity, thus achieving reliable early warning and effectively avoiding workpiece scrapping and production interruptions due to sudden component failure. Energy efficiency refers to the correspondence between "energy consumption" and "cutting efficiency" during CNC machine tool processing (where "energy consumption" is characterized by spindle current and effective cutting power during cutting, and "cutting efficiency" is characterized by tool feed rate). Second, this invention cleverly solves the problem of running complex algorithms in resource-constrained embedded environments. Throughout the embodiment, for example, DMA hardware acceleration and lock-free queues are used in step S1, kurtosis analysis is used instead of complex spectrum analysis in step S3, and lookup table-based methods are employed. Instead of resource-intensive AI models, expert decision matrices are used. These lightweight algorithms, optimized for embedded environments, ensure that the entire monitoring and decision-making process can run efficiently and in real-time on low-cost, low-power microcontrollers, without relying on expensive cloud computing. Finally, this invention significantly improves the reliability and practical value of monitoring. By innovatively combining an energy-efficiency model (reflecting progressive wear) with "vibration morphology" analysis (capturing impact failures) in step S3, the false alarm rate is significantly reduced. More importantly, through intelligent decision-making in step S3 and parameter closed-loop injection in step S4, this invention constructs a complete high-speed closed loop from "detection-diagnosis-decision-control" locally in the embedded system. This enables the system to autonomously adjust operating parameters based on the real-time health status of components, extending component lifespan, maximizing their remaining value, and ensuring maximum processing quality. This completely changes the limitations of traditional monitoring systems that can only passively alarm, giving equipment true autonomous optimization capabilities. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of an embedded monitoring method for tool wear in CNC machine tools. Detailed Implementation

[0014] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0015] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the term "first" may be used herein to describe various units, these units should not be limited by this term. These terms are used merely to distinguish one unit from another. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] To achieve the above objectives, please refer to Figure 1 This invention provides an embedded monitoring method for tool wear in CNC machine tools, comprising the following steps:

[0018] Step S1: Embedded monitoring of the tool during the machining process, synchronous multimodal data acquisition and preprocessing to form a real-time working condition data frame queue; data of the effective cutting range is extracted from the real-time working condition data frame queue to obtain the effective cutting data sequence; the effective cutting data sequence is discretized for current and tool feed rate to form an energy-efficiency reference fingerprint.

[0019] In this embodiment, this step first synchronously acquires multimodal data such as spindle current, rotational speed, feed rate, and vibration through an embedded real-time interrupt. The "instantaneous effective cutting power" is then calculated in a lightweight manner within the interrupt, and this data is encapsulated into a "real-time operating condition data frame" and stored in a queue. Next, in the background analysis task, statistical methods are used to automatically identify and remove idle and unsteady-state cutting data, resulting in a clean "effective cutting data sequence." Finally, the "effective cutting power" in this sequence is discretized into intervals, and the "actual feed rate" corresponding to each interval is averaged, thereby constructing and solidifying an energy-efficiency benchmark fingerprint lookup table describing the standard efficiency of the new tool under different power levels.

[0020] Step S2: Extract the expected healthy feed rate from the energy efficiency benchmark fingerprint; calculate the absolute amount of performance deviation based on the expected healthy feed rate, and perform cutting performance degradation analysis to obtain the degree of cutting performance degradation;

[0021] In this embodiment, this step is executed in a background analysis task. For each real-time working condition data frame in an effective cutting state, the task uses the "instantaneous effective cutting power" as an index to quickly query the pre-established energy-efficiency benchmark fingerprint to obtain an "expected healthy feed rate". Then, the difference between this expected value and the "actual feed rate" contained in the data frame is calculated to obtain an "absolute feed rate deviation" with physical units. To eliminate the inconsistency of evaluation scales under different working conditions, this absolute deviation is normalized and smoothed through an IIR low-pass filter to finally obtain a stable, dimensionless "cutting performance degradation".

[0022] Step S3: Calculate the vibration morphological anomaly degree of the real-time working condition data frame queue to obtain the vibration morphological anomaly degree; perform time-series analysis of tool wear trend based on cutting performance attenuation degree to obtain wear dynamic trend value; determine tool wear based on wear dynamic trend value and vibration morphological anomaly degree, optimize machining quality, and generate optimization strategy type; calculate feed rate adjustment according to optimization strategy type to obtain feed rate dynamic adjustment coefficient.

[0023] In this embodiment, this step is executed within a background decision-making task. First, this task calculates two core diagnostic indicators in parallel: Firstly, by comparing the short-term and long-term averages of historical "cutting performance attenuation," a "wear dynamic trend value" reflecting the rate of change is calculated; secondly, by performing time-domain statistics (e.g., calculating kurtosis) on the raw vibration signal data, a "vibration morphology anomaly degree" reflecting the quality risk of the machined surface is obtained without complex spectral analysis. Then, a lightweight embedded expert system performs multi-objective decision matching on the data from these three dimensions—"cutting performance attenuation," "wear dynamic trend value," and "vibration morphology anomaly degree"—along with an externally configurable "quality assurance priority," selecting the "optimization strategy type" best suited to the current scenario from a pre-defined strategy library. Finally, based on the selected strategy type, a multi-factor coupled calculation formula comprehensively considers the current attenuation level and future deterioration trends, and performs constraint processing on safety boundaries and change slopes to calculate the final "feed rate dynamic adjustment coefficient."

[0024] Step S4: Write the dynamic adjustment coefficient of the feed rate into the CNC controller of the CNC machine tool through the preset communication interface for closed-loop adjustment of machining parameters, and generate closed-loop adjustment instructions for machining parameters.

[0025] In this embodiment, this step is executed in a separate background communication task. This task first identifies the brand and model of the currently connected CNC controller through an automatic detection and polling mechanism, and loads the matching communication protocol, register address, data format, and legal value range from its internal "protocol adaptation template library." Next, it performs boundary trimming, format conversion, and byte order adjustment on the "feed rate dynamic adjustment coefficient" calculated in the previous step according to the loaded template, generating a "formatted control value" that the controller can directly parse. Then, it encapsulates this value into a "parameter control message" conforming to a specific industrial bus protocol, with a checksum and serial number. Finally, it sends this message to the CNC controller through the corresponding physical interface, and verifies whether the parameters were successfully written by reading the receipt or performing a reverse lookup, thus completing a complete and reliable local closed-loop control action from monitoring and diagnosis to autonomous optimization.

[0026] Preferably, step S1 involves embedded monitoring of the cutting tool during the machining process, and synchronous multimodal data acquisition and preprocessing, including:

[0027] The system acquires the trigger signal for the start of the processing task, performs embedded real-time interrupt configuration and resource reservation processing, and forms an embedded sampling interrupt environment. The embedded sampling interrupt environment includes the real-time status of the configured hardware timer, DMA channel and static pre-allocated buffer.

[0028] Efficient synchronous acquisition of multi-channel data is performed based on an embedded sampling interrupt environment to obtain the original sensor dataset.

[0029] The baseline parameter embedded adaptive calibration is performed on the original sensor dataset to obtain the compensated no-load current baseline.

[0030] Based on the original sensor dataset and the compensated no-load current baseline, the power is calculated in real time using embedded optimization to obtain the effective cutting power value;

[0031] Embedded real-time data queue management is implemented for effective cutting power values ​​and raw sensor datasets to form a real-time operating condition data frame queue.

[0032] In one embodiment, after receiving a rising edge signal from the CNC system I / O port as a trigger signal to start the machining task, the microcontroller of the embedded device immediately configures its on-chip hardware timer (TIM2), setting its overload value to 499 and its prescaler value to 71, thereby generating an update interrupt with a period of 1 millisecond. The priority of this update interrupt is set to the highest priority 0 in the system. Simultaneously, the direct memory access (DMA) controller is configured, setting its data transfer direction to from peripheral to memory, its transfer mode to circular mode, and statically pre-allocating a 256-byte memory block as a data receive buffer. The configured hardware timer, DMA channel, and statically pre-allocated buffer together constitute the embedded sampling interrupt environment.

[0033] In each 1-millisecond interrupt service routine (ISR) generated by a hardware timer (TIM2), the DMA controller is automatically triggered, immediately initiating a sequence of acquisitions from the multi-channel analog-to-digital converter (ADC) without microcontroller kernel intervention. This sequence acquisition strictly follows a preset channel order, synchronously acquiring the analog voltage values ​​from the spindle motor current sensor and the X, Y, and Z axis analog voltage values ​​from the triaxial accelerometers rigidly connected to the tool holder. Almost simultaneously, the microcontroller kernel sends a Remote Transfer Request (RTR) frame with ID 0x101 to the servo drivers of the spindle and feed axes via the Controller Area Network (CAN) bus to read the raw digital values ​​of the current rotational speed and current feed rate from their internal registers. All acquired raw sensor readings together constitute the raw sensor dataset.

[0034] During the initial learning phase of the equipment, a separate low-priority task is used to statistically analyze data from 1000 consecutive raw sensor datasets where the spindle motor current value is below a preset coarse threshold (e.g., 2 amps). A frequency histogram of current values ​​across 20 intervals is constructed to identify the interval with the highest frequency, and the arithmetic mean of all current values ​​within that interval is calculated and used as the "initial no-load current mean". During normal operation, this statistical process is repeated every 100 processing cycles, and the "initial no-load current mean" is updated using a first-order IIR low-pass filter. The update method is: new mean = 0.1 × current statistical mean + 0.9 × old mean. The mean after this adaptive update is the compensated no-load current baseline.

[0035] Based on the original sensor dataset and the compensated no-load current baseline, embedded optimization is performed in real-time to calculate the effective cutting power value. This step, within the interrupt service routine, decouples and purifies multiple physical quantities into a single, more fundamental indicator through a lightweight calculation. Specifically, after each acquisition of the original sensor dataset, a subtraction and multiplication operation is immediately performed, calculated as follows: ;in, The effective cutting power value is... The instantaneous current values ​​are from the original sensor dataset. This is the baseline of the compensated no-load current. To read or preset the DC bus voltage value (e.g., 310 volts) via the bus, The preset motor efficiency coefficient (e.g., 0.85).

[0036] The effective cutting power value obtained by this calculation has removed some of the influence of no-load power consumption and speed variation compared to the original current value, and can better represent the net power actually used for material removal.

[0037] The effective cutting power value calculated in the previous step, along with the raw values ​​of the actual feed rate and triaxial acceleration from the original sensor dataset, and a timestamp provided by the counter value of the hardware timer (TIM2), are encapsulated into a fixed-size 24-byte structure as a "real-time operating condition data frame". Subsequently, this data frame is stored in a pre-allocated 4-kilobyte circular buffer through a single memory copy operation, and the write pointer is atomically moved. This circular buffer constitutes the real-time operating condition data frame queue, serving as an asynchronous communication hub between high-priority interrupt tasks and medium-priority analysis tasks. Its lock-free design ensures efficient data transmission and the stability of the real-time system.

[0038] Preferably, step S1, which involves extracting the effective cutting interval data from the real-time operating condition data frame queue, includes:

[0039] The no-load current feature is extracted from the real-time operating condition data frame queue to obtain the no-load current reference value;

[0040] The cutting interval is initially identified by using the no-load current reference value to obtain the original cutting interval set from the real-time operating condition data frame queue.

[0041] The original cutting interval set is corrected for working condition change boundaries to obtain the cutting interval boundaries;

[0042] Based on the boundaries of the cutting intervals, the steady-state cutting segments of the real-time working condition data frame queue are identified to obtain the set of steady-state cutting intervals.

[0043] By traversing the set of steady-state cutting intervals, effective data extraction and merging are performed on the real-time working condition data frame queue to obtain an effective cutting data sequence.

[0044] In another embodiment, at the start of each processing cycle, the analysis task continuously retrieves 1000 data frames from the real-time operating condition data frame queue and extracts the instantaneous current values ​​of the spindle motor contained therein. A frequency histogram of current values ​​covering the range of 0 to 3 amperes with a resolution of 0.1 amperes is constructed to determine the interval with the highest frequency. The arithmetic mean and standard deviation of all current values ​​within this interval are calculated, and these two statistics are used as the reference value for the no-load current.

[0045] Next, the cutting interval is initially identified in the real-time operating condition data frame queue using the no-load current reference value. This process utilizes statistical principles to generate a dynamic threshold to distinguish between cutting and non-cutting operations. Specifically, a cutting initiation threshold is set, with the value of: ;in, The cutting start threshold is used to determine whether the tool has entered an effective cutting state. μ is the mean of the no-load current reference values, and σ is the standard deviation of the no-load current reference values. Each data frame in the real-time operating condition data frame queue is checked one by one. If the instantaneous current value of a data frame first exceeds the cutting start threshold, the timestamp of that data frame is marked as the start of a cutting segment; if the instantaneous current value of a subsequent data frame first falls below the cutting start threshold, the timestamp of that data frame is marked as the end of that cutting segment. All marked start-end timestamp pairs together constitute the original cutting interval set.

[0046] To eliminate interference from transient impacts caused by spindle acceleration / deceleration or workpiece entry / exit from the cutting zone, the boundaries of the initially identified intervals are corrected. Specifically, each interval in the original cutting interval set is traversed, discarding 50 milliseconds of data at its start and end. For example, for an interval from time point t_1 to t_2, the corrected interval is [t_1+50ms, t_2-50ms]. This operation aims to eliminate unstable states at the beginning and end of each cutting segment, ensuring that subsequent analysis is based on steady-state cutting data. All the corrected intervals together constitute the cutting interval boundaries.

[0047] Steady-state cutting segments are identified in the real-time working condition data frame queue based on the cutting interval boundaries. This step aims to further refine the data and identify the intervals truly used for stable machining. Specifically, within each interval defined by the cutting interval boundaries, the mean and standard deviation of the "effective cutting power" values ​​of all data frames within that interval are calculated. If the ratio of the power standard deviation to the mean (i.e., the coefficient of variation) of an interval is less than a preset steady-state discrimination coefficient (e.g., 0.15), then that interval is determined to be a steady-state cutting segment. All intervals determined to be steady-state cutting segments together constitute the cutting steady-state interval set.

[0048] The analysis task extracts only the data frames whose timestamps fall within these steady-state intervals from a large queue of real-time operating condition data frames, based on the time range recorded in the set of steady-state cutting intervals. Then, the data frames extracted from all steady-state cutting segments are merged into a single, continuous data sequence in chronological order. This sequence is the effective cutting data sequence, which provides a clean data source for subsequently constructing a high-precision energy-efficiency benchmark fingerprint.

[0049] Preferably, the discretization of the effective cutting data sequence for current and tool feed rate in step S1 includes:

[0050] The effective cutting data sequence is discretized into power domain intervals and mapped to obtain a discrete interval rate set.

[0051] Perform interval characteristic value statistics on the discrete interval rate set to form interval characteristic values;

[0052] The discrete interval rate set and interval feature values ​​are solidified into an energy-efficiency benchmark fingerprint.

[0053] In one embodiment, a power domain discretization interval mapping is performed on the effective cutting data sequence. Continuous, multi-dimensional machining state data is transformed into a structured, easily queryable discrete model. Specifically, this analysis task first traverses the effective cutting data sequence to determine the maximum value P_max and minimum value P_min of all "effective cutting power" values. Then, the power range [P_min, P_max] is uniformly divided into N (e.g., N=50) non-overlapping power discrete intervals. Each interval is uniquely identified by its starting power value. Next, each data frame in the effective cutting data sequence is traversed again, and based on its "effective cutting power" value, it is assigned to the corresponding power discrete interval. Only the "actual feed rate" value of that data frame is added to a dynamic array associated with that interval. Finally, a set containing N dynamic arrays is formed, which is the discrete interval rate set.

[0054] The analysis task traverses N dynamic arrays within a discrete range of feed rates. For each non-empty dynamic array, the task calculates the arithmetic mean of all "actual feed rate" values ​​within the array. This arithmetic mean is considered the standard performance achievable by a healthy cutting tool within that specific power discrete range. The arithmetic mean calculated from all N dynamic arrays collectively constitutes the characteristic value of the range.

[0055] The discrete interval rate set and interval feature values ​​are solidified into an energy-efficiency benchmark fingerprint. Specifically, the analysis task creates a look-up table structure containing N entries. Each entry in this look-up table uses the starting power value of its corresponding discrete power interval as the "key" and the "interval feature value" (i.e., the average rate) calculated in the previous step that belongs to that interval as the "value". After construction, this complete look-up table data structure is written to the flash memory or electrically erasable programmable read-only memory (EEPROM) of the embedded device for persistent storage in one go through a standard non-volatile memory write interface. This solidified look-up table is the energy-efficiency benchmark fingerprint, which provides an efficient and fast lookup benchmark for subsequent online performance quantization steps.

[0056] Preferably, step S2 includes:

[0057] Extract the current instantaneous current of the spindle motor and the actual feed rate from the real-time operating condition data frame queue to construct instantaneous operating condition data pairs;

[0058] Extract the current instantaneous current value from the instantaneous operating condition data pair and use it as an index to extract the reference rate from the energy-efficiency reference fingerprint to obtain the expected healthy feed rate;

[0059] The absolute deviation of the feed rate is obtained by calculating the absolute amount of the performance deviation from the expected healthy feed rate based on the instantaneous operating condition data.

[0060] The attenuation degree is obtained by normalizing and smoothing the attenuation degree using the absolute deviation of the feed rate and the expected healthy feed rate.

[0061] In one embodiment, the current effective cutting power and actual feed rate are extracted from the real-time operating condition data frame queue. Specifically, during normal machine tool machining, the analysis task continuously retrieves a data frame from the head of the real-time operating condition data frame queue. If the data frame is determined to be within the effective cutting range, the "effective cutting power" value and the "actual feed rate" value are extracted from it, and these two values ​​are combined into a data pair as the instantaneous operating condition data pair.

[0062] The effective cutting power value is extracted from the instantaneous operating condition data pairs and used as an index to extract the benchmark rate for the energy-efficiency benchmark fingerprint. Real-time operating conditions are instantaneously compared with historical health models to establish a performance evaluation benchmark. Specifically, the analysis task extracts the "effective cutting power" value from the instantaneous operating condition data pairs. Then, using this power value as an index, a binary search algorithm is used to quickly locate the power discrete interval containing this power value in the energy-efficiency benchmark fingerprint (a lookup table) stored in non-volatile memory. The corresponding "value" for this interval, i.e., the standard healthy feed rate, is then directly read as the expected healthy feed rate.

[0063] Then, the absolute value of the performance deviation is calculated based on the instantaneous operating condition data and the expected healthy feed rate. Specifically, the "actual feed rate" value is extracted from the instantaneous operating condition data. Subsequently, a subtraction operation is performed: the expected healthy feed rate obtained in the previous step is subtracted from this actual feed rate. If the result is positive, its absolute value is the absolute deviation of the feed rate; if it is negative or zero, the absolute deviation of the feed rate is zero. This deviation value is expressed in millimeters per minute, directly quantifying how much slower the current tool is compared to a healthy tool under the same power consumption.

[0064] To eliminate inconsistencies in evaluation metrics under different cutting powers, normalization is performed. Specifically, a division operation is executed: the absolute deviation of the feed rate is divided by the expected healthy feed rate used as a benchmark, resulting in a dimensionless percentage value. To suppress signal noise, this percentage value is then input into a first-order IIR low-pass filter for smoothing, and its update method is as follows: ;in, The updated smoothed value, This is the currently calculated normalized percentage value. The previous smoothed value, This is a smoothing coefficient (e.g., 0.2).

[0065] The value after normalization and smoothing filtering is the cutting performance degradation degree, which will be updated to a global shared memory area for use by other tasks.

[0066] Preferably, the calculation of the absolute amount of performance deviation from the expected healthy feed rate based on instantaneous operating condition data includes:

[0067] The instantaneous operating condition data and the expected healthy feed rate are synchronized and paired to obtain the feed rate pairing value.

[0068] The original feed difference is obtained by calculating the absolute performance difference based on the feed rate pairing values;

[0069] Based on the original feed difference and instantaneous operating condition data, the operating condition adaptive processing is performed to obtain the operating condition calibration difference;

[0070] Transient fluctuation suppression is performed using the operating condition calibration difference to obtain the steady-state performance difference;

[0071] The physical meaning of the steady-state performance difference and instantaneous operating condition data was verified to obtain the absolute deviation of the feed rate.

[0072] In another embodiment, since both the instantaneous operating condition data pair and the expected healthy feed rate originate from the same "effective cutting power" value at the same moment, this step combines these two data sets that are strictly synchronized in terms of time and operating condition. Specifically, a temporary data structure with two members is constructed: one member is the "actual feed rate" extracted from the instantaneous operating condition data pair, and the other member is the "expected healthy feed rate." This data structure is the feed rate pair value.

[0073] Next, the absolute performance difference is calculated based on the feed rate pairing values. Specifically, a subtraction operation is performed: the "expected healthy feed rate" member is subtracted from the "actual feed rate" member in the feed rate pairing values. The result of this operation is the original feed difference, which directly reflects the instantaneous gap between theoretical performance and actual performance.

[0074] Then, condition adaptation processing is performed based on the original feed difference and instantaneous working condition data. This step dynamically adjusts the tolerance for deviations according to the precision of the machining. Specifically, the analysis task extracts the "effective cutting power" value from the instantaneous working condition data pair. If this power value is lower than a preset "finishing power threshold" (e.g., 500 watts), the current condition is determined to be finishing, and the original feed difference is multiplied by a "finishing sensitivity coefficient" greater than 1 (e.g., 1.5); if the power value is higher than the threshold, the condition is determined to be roughing, and the original feed difference remains unchanged. The difference after this multiplication adjustment is the working condition calibration difference.

[0075] Next, transient fluctuation suppression is performed using the operating condition calibration difference. To filter out drastic jumps in individual data points caused by material hardness or measurement noise, this step smooths the continuously calculated operating condition calibration differences. Specifically, the arithmetic mean of the latest five operating condition calibration differences is calculated using a sliding window containing five data points. This average value is the steady-state performance difference, which is more representative of the average performance deviation over a recent period than a single raw difference.

[0076] Finally, the steady-state performance difference and instantaneous operating condition data are mapped to verify their physical meaning. Since tool wear is a unidirectional performance degradation process, only performance loss has a clear physical meaning. Specifically, the steady-state performance difference is judged: if the value is positive, it indicates that there is a performance loss, and it is directly used as the absolute deviation of the feed rate; if the value is negative or zero, it indicates that the performance has not degraded or even slightly exceeded the limit, and does not correspond to wear in a physical sense, so the absolute deviation of the feed rate is forcibly set to zero. This operation ensures that the final output deviation value is always non-negative, and only has a non-zero value when performance degradation occurs.

[0077] Preferably, the vibration morphological anomaly calculation of the real-time operating condition data frame queue in step S3 includes:

[0078] Vibration data segmentation and buffering are performed on the real-time operating condition data frame queue to obtain vibration segment data blocks;

[0079] Low-computational-cost statistical feature extraction is performed on the segmented vibration data blocks to obtain a vibration statistical feature set;

[0080] Morphological abrupt changes are detected in the vibration statistical feature set to obtain morphological abrupt change features.

[0081] The impact characteristics are enhanced by performing impact feature enhancement processing based on morphological abruptness characteristics and vibration statistical feature sets to obtain enhanced impact characteristics.

[0082] Vibration morphology anomaly degree is calculated based on enhanced impact characteristics.

[0083] In one embodiment, vibration data is segmented and cached in a real-time operating condition data frame queue. Specifically, the analysis task continuously retrieves N (e.g., N=256) data frames determined to be within the effective cutting range from the real-time operating condition data frame queue. Then, only the raw triaxial acceleration values ​​are extracted from these data frames, and their vector magnitudes are calculated to form a resultant acceleration time series containing N points. This series is stored in a dedicated static array as the vibration segmented data block.

[0084] Next, low-computational-cost statistical feature extraction is performed on the vibration segmented data blocks. This step calculates multiple statistical moments simultaneously in a single traversal to achieve maximum computational efficiency. Specifically, by traversing the vibration segmented data block in a single pass, the first four central moments of the data block are calculated simultaneously using a set of accumulator variables. Then, based on these central moments, two core morphological features, namely kurtosis and skewness, are calculated using the following formulas: The formula for calculating kurtosis is:

[0085] ;

[0086] The formula for calculating skewness is:

[0087] ;

[0088] in For ravine, For skewness, It is the fourth-order central moment. The central moment is the third order. The standard deviation is denoted as .

[0089] The calculated kurtosis and skewness values ​​together constitute the vibration statistical feature set. Then, morphological feature abrupt change detection is performed on the vibration statistical feature set. Specifically, the analysis task maintains a historical buffer storing the 10 most recently calculated vibration statistical feature sets. The kurtosis change is obtained by calculating the difference between the current kurtosis value and the average kurtosis value in the historical buffer. If this kurtosis change exceeds a preset abrupt change detection threshold (e.g., an absolute change exceeding 2.0), a morphological abrupt change is determined to have occurred. This determination result (a Boolean value, true or false) is the morphological abrupt change feature.

[0090] Next, impact feature enhancement processing is performed based on morphological abrupt change characteristics and vibration statistical feature sets. This step aims to weight and amplify impact signals indicating early chipping or other faults. Specifically, if the morphological abrupt change characteristic is true, the current kurtosis value in the vibration statistical feature set is multiplied by an impact enhancement coefficient greater than 1 (e.g., 1.5); if the morphological abrupt change characteristic is false, the current kurtosis value remains unchanged. The kurtosis value after this conditional multiplication operation is the enhanced impact feature.

[0091] Finally, the vibration morphology anomaly degree is calculated based on the enhanced impact characteristics. Specifically, the enhanced impact characteristics are compared with the "healthy vibration kurtosis baseline" value recorded in the first learning phase. By calculating the absolute value of the difference between the two and normalizing it, a dimensionless percentage value is obtained, which is the vibration morphology anomaly degree. It comprehensively reflects the severity of the current vibration waveform deviating from a healthy state due to the impact event.

[0092] Preferably, step S3, which involves determining tool wear based on wear dynamic trend values ​​and vibration anomaly levels, and optimizing machining quality, includes:

[0093] The wear dynamic trend value is used to determine the wear level by partitioning the wear level, and the tool wear level indication is obtained;

[0094] Based on the tool wear level indication, vibration anomaly degree, and cutting performance attenuation degree, the impact on machining quality is predicted, and the quality assurance priority is obtained;

[0095] Based on the quality assurance priority and combined with the pre-built optimization strategy library, multi-objective strategy matching is performed to generate optimization strategy types.

[0096] In one embodiment, the wear dynamic trend value is used to determine the wear level. Specifically, the decision task compares the wear dynamic trend value read from the shared memory area with a "wear grading threshold table" stored in memory. This table divides the wear dynamic trend value into four levels: less than 0.05 is level 0 (normal); [0.05, 0.1) is level 1 (light wear); [0.1, 0.2) is level 2 (moderate wear); and greater than or equal to 0.2 is level 3 (severe wear). By performing a range judgment, the current level is determined, and the code (0, 1, 2, 3) of this level is the tool wear level indicator.

[0097] Next, machining quality impact is predicted based on tool wear level indication, vibration anomaly degree, and cutting performance degradation. This step integrates multiple heterogeneous diagnostic indicators to assess the risk to final product quality. Specifically, an initial quality impact score is calculated using the following formula: ;in, To score the impact on quality, For the degree of reduction in cutting performance, For vibration morphology anomaly, and Preset weighting coefficients (e.g.) =0.6, =0.4), and The impact coefficient is retrieved from the built-in lookup table based on the current processing material and process type. Then, a skip-level judgment rule is added: if the tool wear level indicator is 1 (slight), but the vibration morphology anomaly instantaneously exceeds a high-risk threshold (e.g., 0.8), the quality impact score is forcibly set to the highest score of 10. The final score of 0 to 10 is the quality assurance priority.

[0098] Finally, multi-objective strategy matching is performed based on the quality assurance priority and a pre-built optimization strategy library. This step uses a lightweight expert decision-making system to map multi-dimensional inputs to a specific action strategy. Specifically, the tool wear level indication, quality assurance priority, and a "trend state" obtained by discretizing the wear dynamic trend value are combined into a decision vector. Then, this vector is used to query an "N-dimensional expert decision matrix" stored in memory. Each cell of this matrix pre-stores an code representing a specific optimization strategy (e.g., 0x01 represents an efficiency-first strategy, 0x02 represents a quality-first strategy, 0x03 represents a lifespan extension strategy, and 0x04 represents a protection and preservation strategy). For example, when the decision vector is [Level 2, Priority > 7, Trend Deterioration], the query result is 0x04. This code retrieved from the matrix is ​​the optimization strategy type.

[0099] Preferably, step S3, which involves calculating the feed rate adjustment based on the optimization strategy type, includes:

[0100] Calculate the basic adjustment coefficient based on the optimization strategy type and the degree of cutting performance degradation;

[0101] Analyze the magnitude and direction of the dynamic wear trend value, and perform trend response compensation on the basic adjustment coefficient to form the trend-compensated coefficient.

[0102] Obtain the current process parameter information, combine it with the trend compensation coefficient to make process adaptability adjustments, and obtain the adjusted process coefficients;

[0103] Boundary constraint processing is performed using the process adjustment coefficients to obtain the constraint adjustment coefficients;

[0104] The feed rate dynamic adjustment coefficient is obtained by performing time-domain smoothing on the constraint adjustment coefficient.

[0105] In one embodiment, a base adjustment coefficient is calculated based on the optimization strategy type and the degree of cutting performance degradation. Specifically, this decision-making task uses the optimization strategy type encoding to query the "base degradation response coefficient" k_base corresponding to the strategy from a built-in "strategy calculation parameter table". Then, a preliminary adjustment value reflecting the current wear state is calculated using a linear formula, the calculation method of which is as follows: ;in, This refers to the basic adjustment coefficient. This represents the degree of reduction in cutting performance.

[0106] Next, the magnitude and direction of the wear dynamic trend value are analyzed, and trend response compensation is applied to the basic adjustment coefficient. This step introduces foresight into the control system, allowing it to respond to the rate of wear deterioration. Specifically, the "trend response coefficient" corresponding to the current strategy is retrieved from the "strategy calculation parameter table". Then, the trend effect is coupled in using the following formula:

[0107] ;

[0108] in, The coefficient after trend compensation. This represents the wear change rate extracted from the wear dynamic trend value. This calculation enables predictive compensation for future wear. Then, current process parameter information is obtained, and process adaptive adjustments are made based on the trend-compensated coefficients. Specifically, the decision-making task reads the current machining condition identified in the first step from the shared memory area, such as "roughing" or "finishing". If the current condition is "roughing", the trend-compensated coefficients are multiplied by a "roughing tolerance coefficient" greater than 1 (e.g., 1.1) to allow for higher efficiency in non-critical machining stages; if the condition is "finishing", it remains unchanged. The value after this adaptive multiplication operation is the process adjustment coefficient.

[0109] Next, boundary constraint processing is performed using the adjusted process coefficients. To ensure safety, this step clamps the calculated coefficient values. Specifically, the adjusted process coefficients are compared with a preset "absolute safety range" (e.g., [0.5, 1.0]). If the calculated value is lower than 0.5, it is forcibly set to 0.5; if it is higher than 1.0, it is forcibly set to 1.0. This operation ensures that the output commands are always within the acceptable safety boundaries of the machine tool. The value after this boundary constraint processing is the constrained adjustment coefficient.

[0110] Finally, the constraint-adjusted coefficients are smoothed in the time domain. To avoid drastic changes in control commands impacting the machine tool, this step limits the slope of the output value. Specifically, the difference between the constraint-adjusted coefficient and the previous final output value is calculated. If the absolute value of this difference exceeds a preset "maximum change slope" (e.g., a change of no more than 0.05 per 100 milliseconds), the current output value will be limited so that its change is exactly equal to this maximum slope. The value after this smoothing process is the feed rate dynamic adjustment coefficient, which will be used as the final control quantity and transmitted to the next step.

[0111] Preferably, step S4 includes:

[0112] Perform status detection on the preset communication interface to obtain communication link status information;

[0113] Based on the communication link status information and the dynamic adjustment coefficient of the feed rate, the CNC controller compatibility is analyzed to obtain the controller parameter mapping information;

[0114] Based on the controller parameter mapping information, the dynamic adjustment coefficient of the feed rate is converted into a command data format to obtain a formatted control value;

[0115] Parameter control messages are constructed based on formatted control values ​​and communication link status information;

[0116] The parameter control message is injected, executed, and verified to obtain the closed-loop adjustment command for the processing parameters.

[0117] In one embodiment, a status check is performed on a preset communication interface. Specifically, when the communication task starts, it checks the physical connection status sequentially according to a preset priority list (e.g., EtherCAT, Profinet, Modbus TCP, CANopen). This is done by sending a protocol-specific link probe message (e.g., an EtherCAT link status query frame) and waiting for a valid response. The first interface to receive a valid response is determined as the currently active interface. Information such as the interface type and the measured response time is recorded as the communication link status information.

[0118] Next, CNC controller compatibility analysis is performed based on communication link status information and feed rate dynamic adjustment coefficients. This step enables plug-and-play functionality and adaptability to different brand controllers. Specifically, after determining the active interface, the communication task sends a standard vendor and device ID read command through that interface. Based on the returned ID information, a template matching that ID is queried and loaded from a "controller protocol adaptation template library" stored in memory. This template precisely defines the register address, data type, valid value range, and byte order corresponding to the "feed rate adjustment" function. These parameters retrieved from the template together constitute the controller parameter mapping information.

[0119] Then, the feed rate dynamic adjustment coefficient is converted into an instruction data format based on the controller parameter mapping information. Specifically, the communication task converts the logical percentage form of the feed rate dynamic adjustment coefficient into a physical data format that the target controller can recognize. For example, if the data type defined in the controller parameter mapping information is a 16-bit unsigned integer with a range of 0-1000 corresponding to 0-100%, then multiplication and rounding operations are performed: encoded value = round(feed rate dynamic adjustment coefficient × 1000). If the defined byte order is big-endian, the high and low bytes of the 16-bit encoded value are also swapped. The value after this conversion and byte order adjustment is the formatted control value.

[0120] Next, a parameter control message is constructed based on the formatted control values ​​and communication link status information. Specifically, the communication task calls the corresponding message construction function according to the protocol type determined in the communication link status information. For example, for the Modbus TCP protocol, the task will construct a standard Modbus message, where the function code is set to 0x06 (write a single register), the register address is set to the address obtained from the controller parameter mapping information, and the data written is the formatted control value generated in the previous step. The message's transaction identifier is set to an auto-incrementing sequence number, and a message header is calculated and appended. The completed byte stream is the parameter control message.

[0121] Finally, parameter injection and verification are performed on the parameter control message. Specifically, the communication task sends the parameter control message through a selected physical interface (such as an Ethernet port) and starts a timeout timer (e.g., 50 milliseconds). If a correct response message is received from the controller before the timeout, the write is considered successful. For verification, the task immediately sends a request to read the same register address and compares the returned value with the sent value. If they match, the parameter is confirmed to be effective. A complete transmission and verification record containing the target address, transmitted data, timestamp, and success status constitutes the closed-loop adjustment instruction for the processing parameters.

[0122] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0123] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. An embedded monitoring method for tool wear in CNC machine tools, characterized in that, Includes the following steps: Step S1: Embedded monitoring of the cutting tool during the machining process, synchronous multimodal data acquisition and preprocessing, forming a real-time working condition data frame queue; The effective cutting data sequence is obtained by extracting the effective cutting data from the real-time working condition data frame queue. Discretize the effective cutting data sequence with current and tool feed rate to form an energy-efficiency reference fingerprint; Step S2: Extract the expected healthy feed rate from the energy efficiency benchmark fingerprint; The absolute value of the performance deviation is calculated based on the expected healthy feed rate, and the degree of cutting performance degradation is analyzed to obtain the degree of cutting performance degradation. Step S3: Calculate the vibration morphological anomaly degree of the real-time operating condition data frame queue to obtain the vibration morphological anomaly degree; Based on the degree of cutting performance decay, the wear trend of the tool is analyzed over time to obtain the dynamic wear trend value; Tool wear is determined based on wear dynamic trend value and vibration anomaly degree, and machining quality is optimized to generate optimization strategy type; feed rate adjustment is calculated according to optimization strategy type to obtain feed rate dynamic adjustment coefficient; Step S4: Write the dynamic adjustment coefficient of the feed rate into the CNC controller of the CNC machine tool through the preset communication interface for closed-loop adjustment of machining parameters, and generate closed-loop adjustment instructions for machining parameters. In step S1, the embedded monitoring of the tool during the machining process and the synchronous multimodal data acquisition and preprocessing are specifically as follows: the trigger signal for the start of the machining task is obtained, embedded real-time interrupt configuration and resource reservation processing are performed to form an embedded sampling interrupt environment, wherein the embedded sampling interrupt environment includes the real-time status of the configured hardware timer, DMA channel and static pre-allocated buffer. Efficient synchronous acquisition of multi-channel data is performed based on an embedded sampling interrupt environment to obtain the original sensor dataset; the baseline parameters of the original sensor dataset are then subjected to embedded adaptive calibration to obtain the compensated no-load current baseline. Based on the original sensor dataset and the compensated no-load current baseline, the power is calculated in real time using embedded optimization to obtain the effective cutting power value; Embedded real-time data queue management is implemented for effective cutting power values ​​and raw sensor datasets to form a real-time operating condition data frame queue.

2. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S1, which involves extracting the effective cutting interval data from the real-time operating condition data frame queue, includes: The no-load current feature is extracted from the real-time operating condition data frame queue to obtain the no-load current reference value; The cutting interval is initially identified by using the no-load current reference value to obtain the original cutting interval set from the real-time operating condition data frame queue. The original cutting interval set is corrected for working condition change boundaries to obtain the cutting interval boundaries; Based on the boundaries of the cutting intervals, the steady-state cutting segments of the real-time working condition data frame queue are identified to obtain the set of steady-state cutting intervals. By traversing the set of steady-state cutting intervals, effective data extraction and merging are performed on the real-time working condition data frame queue to obtain an effective cutting data sequence.

3. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S1 involves discretizing the effective cutting data sequence for current and tool feed rate, including: The effective cutting data sequence is discretized into power domain intervals and mapped to obtain a discrete interval rate set. Perform interval characteristic value statistics on the discrete interval rate set to form interval characteristic values; The discrete interval rate set and interval feature values ​​are solidified into an energy-efficiency benchmark fingerprint.

4. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S2 includes: Extract the current instantaneous current of the spindle motor and the actual feed rate from the real-time operating condition data frame queue to construct instantaneous operating condition data pairs; Extract the current instantaneous current value from the instantaneous operating condition data pair and use it as an index to extract the reference rate from the energy-efficiency reference fingerprint to obtain the expected healthy feed rate; The absolute deviation of the feed rate is obtained by calculating the absolute amount of the performance deviation from the expected healthy feed rate based on the instantaneous operating condition data. The attenuation degree is obtained by normalizing and smoothing the attenuation degree using the absolute deviation of the feed rate and the expected healthy feed rate.

5. The embedded monitoring method for CNC machine tool tool wear according to claim 4, characterized in that, The calculation of the absolute value of the performance deviation from the expected healthy feed rate based on instantaneous operating data includes: The instantaneous operating condition data and the expected healthy feed rate are synchronized and paired to obtain the feed rate pairing value. The original feed difference is obtained by calculating the absolute performance difference based on the feed rate pairing value; Based on the original feed difference and instantaneous operating condition data, the operating condition adaptive processing is performed to obtain the operating condition calibration difference; Transient fluctuation suppression is performed using the operating condition calibration difference to obtain the steady-state performance difference; The physical meaning of the steady-state performance difference and instantaneous operating condition data was verified to obtain the absolute deviation of the feed rate.

6. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S3, which involves calculating the vibration morphological anomaly degree of the real-time operating condition data frame queue, includes: Vibration data segmentation and buffering are performed on the real-time operating condition data frame queue to obtain vibration segment data blocks; Low-computational-cost statistical feature extraction is performed on the segmented vibration data blocks to obtain a vibration statistical feature set; Morphological abrupt changes are detected in the vibration statistical feature set to obtain morphological abrupt change features. The impact characteristics are enhanced by performing impact feature enhancement processing based on morphological abruptness characteristics and vibration statistical feature sets to obtain enhanced impact characteristics. Vibration morphology anomaly degree is calculated based on enhanced impact characteristics.

7. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S3 involves determining tool wear based on wear dynamic trend values ​​and vibration anomaly levels, and optimizing machining quality, including: The wear dynamic trend value is used to determine the wear level by partitioning the wear level, and the tool wear level indication is obtained; Based on the tool wear level indication, vibration anomaly degree, and cutting performance attenuation degree, the impact on machining quality is predicted, and the quality assurance priority is obtained. Based on the quality assurance priority and combined with a pre-built optimization strategy library, multi-objective strategy matching is performed to generate optimization strategy types.

8. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S3, which involves calculating the feed rate adjustment based on the optimization strategy type, includes: Calculate the basic adjustment coefficient based on the optimization strategy type and the degree of cutting performance degradation; Analyze the magnitude and direction of the dynamic wear trend value, and perform trend response compensation on the basic adjustment coefficient to form the trend-compensated coefficient. Obtain the current process parameter information, combine it with the trend compensation coefficient to make process adaptability adjustments, and obtain the adjusted process coefficients; Boundary constraint processing is performed using the process adjustment coefficients to obtain the constraint adjustment coefficients; The feed rate dynamic adjustment coefficient is obtained by performing time-domain smoothing on the constraint adjustment coefficient.

9. The embedded monitoring method for CNC machine tool tool wear according to claim 1, characterized in that, Step S4 includes: Perform status detection on the preset communication interface to obtain communication link status information; Based on the communication link status information and the dynamic adjustment coefficient of the feed rate, the CNC controller compatibility is analyzed to obtain the controller parameter mapping information; Based on the controller parameter mapping information, the dynamic adjustment coefficient of the feed rate is converted into a command data format to obtain a formatted control value; Parameter control messages are constructed based on formatted control values ​​and communication link status information; The parameter control message is injected, executed, and verified to obtain the closed-loop adjustment command for the processing parameters.

Images