A method and system for real-time monitoring of production status based on production line data aggregation

Abstract

This invention relates to the field of industrial control technology, and in particular to a method and system for real-time monitoring of production status based on production line data aggregation. The method includes: real-time acquisition of spindle current, spindle command speed, and workpiece material hardness coefficient; constructing a theoretical energy consumption benchmark based on pre-calibrated equipment dynamic parameters and spindle command speed; adaptively compensating for the difference between the spindle current and the theoretical energy consumption benchmark, combined with the workpiece material hardness coefficient, to calculate the state response intensity; normalizing the state response intensity based on a dynamically determined sliding window of the signal spectrum to generate normalized state features; and inputting the normalized state features into a pre-trained LSTM model to identify the current production status of the production line. This invention improves the real-time performance and accuracy of production line monitoring by constructing a theoretical energy consumption benchmark to separate operating conditions, extracting features from the state response intensity, and dynamically normalizing the spectrum to enhance consistency, improve adaptability to multiple product types, avoid model retraining, and enhance the real-time performance and accuracy of production line monitoring.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and in particular to a method and system for real-time monitoring of production status based on production line data aggregation. Background Technology

[0002] In modern discrete manufacturing production lines, real-time monitoring of production status is the core of achieving digital management. Production line equipment generates massive amounts of underlying physical data, such as current, voltage, and vibration. However, management requires production logic states with clear semantics, such as waiting for materials, roughing, finishing, idling, and tool wear. To break down the information barriers between underlying physical data and production logic states, and to achieve effective conversion from raw sensor data to business decision states, existing technologies typically employ deep learning methods based on Long Short-Term Memory (LSTM) networks. Leveraging the advantages of LSTM in processing long-sequence data, an end-to-end mapping model from physical data to production states is established.

[0003] However, traditional LSTM-based monitoring methods suffer from poor generalization ability to changes in operating conditions. Industrial sites often face multi-variety, small-batch production modes, which means that the processing parameters of production line equipment, such as spindle speed, feed rate, and depth of cut, as well as the processing objects, such as materials of different hardness, will frequently change. Traditional LSTM models are trained under specific historical operating condition data distributions. When the production line switches to new operating conditions, the baseline values ​​and fluctuation patterns of physical data will undergo significant physical drift. For example, the normal vibration amplitude at high speed may be much larger than the fault vibration amplitude at low speed. The difference in distribution can cause the fixed-weight LSTM model to misjudge, misidentifying normal operating condition adjustments as abnormal or incorrect production states. To solve the misidentification problem, it is usually necessary to recollect data and retrain the model for the new operating conditions. However, retraining will produce significant time delays and resource consumption, thus affecting the real-time performance and accuracy of production line status monitoring. Summary of the Invention

[0004] To address the issues of poor model generalization ability and inability to adapt to changes in operating conditions, which lead to misjudgments and affect the real-time performance and accuracy of production status monitoring in traditional LSTM models, this invention provides a method and system for real-time monitoring of production status based on production line data aggregation.

[0005] Firstly, the present invention provides a method for real-time monitoring of production status based on production line data aggregation, employing the following technical solution:

[0006] A real-time production status monitoring method based on production line data aggregation includes: acquiring production line data in real time, including spindle current, spindle command speed, and workpiece material hardness coefficient; constructing a theoretical energy consumption benchmark reflecting the basic energy consumption under the current operating condition of the equipment using the spindle command speed based on pre-calibrated equipment dynamic parameters, wherein the theoretical energy consumption benchmark is composed of a speed-related steady-state energy consumption component and an acceleration-related dynamic energy consumption component; performing differential processing on the spindle current and the theoretical energy consumption benchmark, and performing adaptive compensation in combination with the workpiece material hardness coefficient to calculate the state response intensity, wherein the state response intensity characterizes the operational characteristics after removing interference from operating conditions and material properties; normalizing the state response intensity based on a sliding window dynamically determined by the signal spectrum to generate normalized state features; and inputting the normalized state features into a pre-trained long short-term memory network model to identify the current production status of the production line, thereby completing real-time production status monitoring based on production line data aggregation.

[0007] The beneficial effects are as follows: By constructing a theoretical energy consumption benchmark, the influence of operating condition changes on physical data is effectively eliminated. The steady-state energy consumption and dynamic energy consumption components related to rotational speed are comprehensively considered, ensuring the comprehensiveness of energy consumption assessment. The adaptive compensation mechanism, combined with the material hardness coefficient, further eliminates the interference of material characteristic differences. Through the calculation of state response intensity, the characteristics of business actions are extracted more accurately, eliminating the interference of operating conditions and material characteristics, enabling the LSTM model to focus on identifying real production state changes. The normalization processing based on the dynamic determination of the sliding window according to the signal spectrum improves the consistency of features, enhances the model's adaptability to multi-variety, small-batch production modes, avoids the need for frequent retraining, and improves the real-time performance and accuracy of production status monitoring.

[0008] Furthermore, the theoretical energy consumption benchmark satisfies:

[0009] In the formula, For the current moment The theoretical energy consumption benchmark, This refers to the steady-state energy consumption component related to rotational speed. For acceleration-related dynamic energy dissipation components, For the current moment spindle command speed, For the current moment The difference between the spindle speed commanded at the previous moment and the speed commanded at the previous moment. The sampling interval is... The frictional resistance correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is... The moment of inertia correlation coefficient, which is a pre-calibrated dynamic parameter of the equipment and reflects the relationship between acceleration and energy consumption.

[0010] The beneficial effects are as follows: by constructing a composite function containing a rotational speed square term and an acceleration term, a scientific assessment of the theoretical energy consumption benchmark is achieved. The rotational speed square term accurately reflects the energy consumption characteristics under steady-state operation, while the acceleration term reflects the energy consumption changes during dynamic processes. This effectively isolates the influence of operating condition changes on physical data and provides a reliable energy consumption benchmark for the calculation of state response intensity.

[0011] Furthermore, the calibration method for the equipment dynamic parameters is as follows: when the production line is in an unloaded operating state, the spindle current and spindle command speed are collected synchronously; based on the collected data, the least squares method is used to fit the equation. Thus determine and The numerical value; in the formula, Main spindle current, Main spindle speed command This is the difference between the spindle command speed at the sampling time and the previous sampling time, used to calibrate the equipment's dynamic parameters. The sampling interval is... The frictional resistance correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is... The moment of inertia correlation coefficient, which is a pre-calibrated dynamic parameter of the equipment and reflects the relationship between acceleration and energy consumption.

[0012] Furthermore, the state response strength satisfies:

[0013] In the formula, For the current moment State response intensity, For the current moment spindle current, The rated line voltage on the input side of the spindle drive. For the current moment The theoretical energy consumption benchmark, The current moment is dynamically determined through online learning. The influence coefficient of material hardness, For the current moment The corresponding workpiece material hardness coefficient, The hyperparameter constant is used to prevent the denominator from being zero. It is the natural logarithm function.

[0014] The beneficial effects are as follows: by constructing a logarithmic function that includes the ratio of spindle current to theoretical energy consumption benchmark, the assessment of state response intensity is realized, the influence of operating condition changes is eliminated, online learning of the material hardness influence coefficient ensures adaptive compensation for different material properties, and the natural logarithmic term realizes nonlinear compression of the power ratio, effectively suppressing the influence of extreme values, and providing stable feature input for production state identification.

[0015] Furthermore, the hardness influence coefficient is a dynamically adjusted parameter, and the adjustment method is as follows: when a cutting state is detected, recent data are collected. The state response intensity at each cutting moment and the corresponding workpiece material hardness coefficient; calculation The average ratio of the state response intensity to the corresponding workpiece material hardness coefficient at each cutting moment is used as the material hardness influence coefficient at the current moment.

[0016] Furthermore, it also includes initialization and maintenance processing of the hardness influence coefficient: during system initialization, a fixed initial value is assigned to the hardness influence coefficient from the material database; when the equipment is in a non-cutting state, the hardness influence coefficient is updated using an exponential decay method until it reaches the set lower limit value.

[0017] Furthermore, the normalized state characteristics satisfy:

[0018] In the formula, For the current moment The normalized state characteristics, For the current moment State response intensity, For a moment State response intensity, The length of the sliding window. For the current moment The mean of the state response intensity at all times within the sliding window. The noise tolerance is set.

[0019] The beneficial effects are as follows: by constructing a normalized model that includes state response intensity and sliding window variance, a scientific evaluation of normalized state characteristics is achieved. The state response intensity is standardized as a relative value relative to local fluctuations, effectively eliminating the influence of amplitude differences under different working conditions. The noise tolerance ensures the numerical stability of the denominator, and improves the consistency of features and the recognition accuracy of the LSTM model.

[0020] Furthermore, the noise tolerance is obtained by: collecting the state response intensity for a corresponding duration under the no-load condition of the production line; calculating the standard deviation of the state response intensity under the no-load condition; and taking 1.5 times the standard deviation as the noise tolerance.

[0021] Furthermore, the step of identifying the current production state of the production line includes: inputting normalized state features from multiple consecutive time points, including the current time, into a long short-term memory network model, and processing the probability vector output by the long short-term memory network model. Normalization processing; a majority voting mechanism is used to smooth the prediction results at multiple consecutive time points to determine the current production status of the production line.

[0022] The beneficial effect is that by inputting the normalized state features from multiple consecutive time points into the LSTM model, the correlation of temporal information is fully utilized. Normalization ensures the effectiveness of the output probability vector; the majority voting mechanism smooths the continuous prediction results, effectively reducing the impact of single-point misjudgment and improving the stability and accuracy of production status identification.

[0023] Secondly, this invention provides a real-time production status monitoring system based on production line data aggregation, employing the following technical solution:

[0024] A real-time production status monitoring system based on production line data aggregation includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned real-time production status monitoring method based on production line data aggregation is implemented.

[0025] By adopting the above technical solution, a computer program is generated from the above-mentioned method for real-time monitoring of production status based on production line data aggregation, and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, making them convenient to use.

[0026] The present invention has the following technical effects:

[0027] (1) To address the problem of physical data drift and poor generalization ability caused by changes in working conditions, such as spindle speed adjustment and changes in material properties, such as differences in workpiece hardness, the present invention constructs a steady-state and dynamic dual-component theoretical energy consumption benchmark to more accurately obtain the basic energy consumption characteristics of the equipment under different speeds. Then, it combines the workpiece material hardness coefficient for adaptive compensation to remove working condition interference in the spindle current, such as the energy consumption difference between high speed and low speed, and material interference, such as the difference in cutting load between hard materials and soft materials. The final state response intensity only represents the pure production action characteristics such as waiting for materials, processing, and tool wear, which can more stably reflect the essential attributes of the production state, effectively solve the problem of misjudgment when the working conditions change in the traditional model, and improve the generalization ability of the LSTM model in multi-variety and small-batch production mode.

[0028] (2) Breaking through the limitations of traditional LSTM models that rely on fixed feature preprocessing and need to be retrained for new working conditions, this invention dynamically determines the sliding window based on the signal spectrum, and can autonomously adapt to the differences in the spectrum distribution of physical signals under different working conditions, such as high-frequency vibration signals of high-speed cutting and low-frequency signals of low-speed machining. The state response intensity is normalized through the dynamic window to ensure that the normalized state features input to the LSTM always maintain a stable distribution. This allows the pre-trained LSTM model to directly adapt to the flexible switching needs of multi-variety and small-batch production without having to collect data, adjust parameters or train again for new working conditions. This avoids the time delay and resource consumption caused by the traditional retraining mode and ensures the real-time monitoring of production line status.

[0029] (3) Traditional methods have difficulty in effectively associating the underlying physical data such as spindle current and speed with business logic states such as waiting for materials, roughing, and tool wear. However, the state response intensity obtained by the present invention through theoretical energy consumption benchmark difference and hardness compensation has been stripped of irrelevant interference and focuses only on the core production action features in the production process. After dynamic normalization, the feature recognition is improved. After inputting into the pre-trained LSTM model, it can more accurately identify the current production logic state of the production line, avoid misjudging normal working condition adjustments, such as speed increase, as abnormal or incorrect production states, effectively solve the identification deviation problem caused by feature confusion in traditional models, realize the efficient conversion from underlying perception data to clear business decision state, and improve the accuracy of production state identification. Attached Figure Description

[0030] Figure 1 This is a flowchart of a real-time production status monitoring method based on production line data aggregation according to an embodiment of the present invention.

[0031] Figure 2 This is a schematic diagram showing the time-series comparison between measured spindle current and theoretical energy consumption benchmark in a real-time monitoring method for production status based on production line data aggregation according to an embodiment of the present invention.

[0032] Figure 3 This is a schematic diagram illustrating the temporal variation distribution of the state response intensity in a real-time monitoring method for production status based on production line data aggregation, according to an embodiment of the present invention.

[0033] Figure 4 This is a schematic diagram of the temporal variation distribution of normalized state characteristics in a real-time monitoring method for production status based on production line data aggregation according to an embodiment of the present invention. Detailed Implementation

[0034] 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, not all, of the embodiments of the present invention. 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.

[0035] This invention discloses a method for real-time monitoring of production status based on production line data aggregation, referring to... Figure 1 This includes steps S001-S005:

[0036] S001: Real-time acquisition of production line data including spindle current, spindle command speed, and workpiece material hardness coefficient.

[0037] Specifically, the following data is collected and aggregated in real time through edge gateways deployed on the production line:

[0038] Spindle current: directly acquired by a current sensor, unit: This reflects the real-time energy consumption of the production line; spindle command speed: directly obtained from the PLC system, unit: The data reflects the current processing parameters; the workpiece material hardness coefficient is obtained directly from the MES system or process database, is dimensionless, and reflects the characteristics of the current processing object; the above data are aligned using a timestamp synchronization algorithm.

[0039] S002: Based on the pre-calibrated equipment dynamic parameters, a theoretical energy consumption benchmark reflecting the basic energy consumption under the current operating conditions of the equipment is constructed using the spindle command speed. The theoretical energy consumption benchmark is composed of the steady-state energy consumption component related to the rotation speed and the dynamic energy consumption component related to the acceleration.

[0040] It should be noted that the measured spindle current contains two core components: first, the basic energy consumption determined by the operating conditions of the production line equipment itself, such as speed and acceleration / deceleration; and second, the additional energy consumption generated by production actions, such as cutting and waiting for materials. If the raw current data is used directly, changes in operating conditions will cause fluctuations in the basic energy consumption, masking the characteristics of production actions. Therefore, this step needs to construct a theoretical energy consumption benchmark determined solely by operating parameters based on the principles of equipment dynamics, thus eliminating interference from basic energy consumption and laying the foundation for extracting production-related characteristics.

[0041] Specifically, the calibration method for the device's dynamic parameters is as follows:

[0042] When the production line is in an unloaded state, for example, 5-10 minutes of unloaded operation, the spindle current and spindle command speed are collected synchronously;

[0043] Based on the collected data, the least squares method is used to fit the equation. Thus determine and The value;

[0044] In the formula, Main spindle current, Main spindle speed command This is the difference between the spindle command speed at the sampling time and the previous sampling time, used to calibrate the equipment's dynamic parameters. For example, the sampling interval is... , The frictional resistance correlation coefficient, used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is expressed in units of... , The moment of inertia correlation coefficient, a pre-calibrated dynamic parameter of the equipment, reflects the relationship between acceleration and energy consumption. (Unit: [missing information]) According to the principles of rotating system dynamics, the energy consumption of equipment can be decomposed into two parts: one is the steady-state frictional resistance energy consumption related to the square of the rotational speed, and the other is the dynamic inertial energy consumption related to angular acceleration. This model uses... and The parameters characterize these two physical properties respectively, ensuring that the theoretical energy consumption benchmark is consistent with the actual physical laws.

[0045] Specifically, the theoretical energy consumption benchmark satisfies:

[0046] ;

[0047] In the formula, For the current moment The theoretical energy consumption benchmark, in units of , This refers to the steady-state energy consumption component related to rotational speed. For acceleration-related dynamic energy dissipation components, For the current moment The rate of change of rotational speed, in units of , For the current moment spindle command speed, For the current moment The difference between the spindle speed commanded at the previous moment and the speed commanded at the previous moment. The sampling interval is... The frictional resistance correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is... The moment of inertia correlation coefficient, which is a pre-calibrated dynamic parameter of the equipment and reflects the relationship between acceleration and energy consumption.

[0048] in, This reflects the basic energy consumption required for the production line to maintain a constant speed at the current rotational speed. The higher the rotational speed, the larger this component becomes, indicating that the energy required to overcome mechanical resistance during the constant-speed operation of the production line is higher, which in turn leads to... The larger; This reflects the additional energy consumption required to overcome inertia during acceleration and deceleration of the production line. The greater the rate of change of speed, such as faster acceleration or more abrupt deceleration, the larger this component, indicating a higher dynamic energy consumption demand during operating condition switching of the production line. This will also lead to… The two factors together constitute the theoretical energy consumption benchmark of the production line under no-load conditions, which is determined solely by the operating parameters. This comprehensively covers the impact of both steady-state speed and dynamic acceleration / deceleration on the basic energy consumption.

[0049] like Figure 2 As shown, the horizontal axis represents time and the vertical axis represents current amplitude. The theoretical energy consumption benchmark can accurately follow the steady-state and dynamic changes of the spindle speed, and more accurately depict the basic energy consumption of the equipment determined only by the operating parameters. The difference between the measured current and the theoretical energy consumption benchmark is the additional energy consumption caused by production actions and material hardness, which provides an intuitive basis for subsequent isolation of operating interference and extraction of production status characteristics, and verifies the scientific nature of the theoretical energy consumption benchmark construction.

[0050] S003: The spindle current is differentially processed with the theoretical energy consumption benchmark, and adaptive compensation is performed in combination with the workpiece material hardness coefficient to calculate the state response intensity. The state response intensity characterizes the operational characteristics after the peeling condition and material property interference.

[0051] It should be noted that in actual production, the measured current also includes production actions, such as the additional energy consumption generated by cutting, and this additional energy consumption is affected by the material hardness; for example, cutting loads are greater for high-hardness materials. Therefore, this step needs to combine the measured current with the theoretical energy consumption benchmark and introduce an adaptive compensation term for material hardness to eliminate the dual interference of operating conditions and material characteristics, thereby obtaining the state response characteristics purely caused by production actions and achieving decoupling of business state from the external environment.

[0052] Specifically, the hardness influence coefficient is obtained using the following method:

[0053] When a cutting state is detected, the system automatically collects recent cutting data;

[0054] calculate , The number of valid cutting samples;

[0055] During system initialization, Set to a typical value preset in the material database, such as when processing steel. When the equipment is unloaded, The memory is maintained by an exponential decay method with a decay coefficient of 0.95 to avoid compensation interruption caused by a complete return to zero.

[0056] Specifically, the state response strength satisfies:

[0057] ;

[0058] In the formula, For the current moment State response intensity, For the current moment spindle current, The rated line voltage on the input side of the spindle drive, in units of... , For the current moment The theoretical energy consumption benchmark, The current moment is dynamically determined through online learning. The material hardness influence coefficient, in units of Hardness unit This indicates the additional cutting power corresponding to each unit of hardness coefficient. It represents how many extra watts of power the cutting process consumes when the material hardness increases by one unit. Hardness unit, reflecting the additional energy consumed during cutting for every 1 increase in hardness coefficient. Power; For the current moment The corresponding workpiece material hardness coefficient, This is a hyperparameter constant used to prevent the denominator from being zero; its unit is 1. Exemplary , It is the natural logarithm function.

[0059] in, Essentially, it is the normalized production state response value after eliminating interference from working conditions and material hardness. Its numerical characteristics are directly related to production actions: during rough machining, the cutting load is large, and the measured energy consumption is higher than the benchmark value. It exhibits stable positive values; when operating under no-load conditions while awaiting materials, the measured energy consumption is close to the benchmark value. Approaching zero; cutting resistance increases abnormally when the tool wears. Continuous fluctuations deviating from the normal cutting range will occur; logarithmic transformation not only compresses the dynamic range of the data and reduces the impact of extreme values ​​on subsequent models, but also provides a unified numerical scale for the operational state response under different working conditions and materials, providing crucial support for cross-working-condition feature transfer. In the no-load scenario, and The relational expression is simplified to ,at this time, Approaching zero; in cutting scenarios: By adaptively learning from historical data, it automatically compensates for the influence of material hardness, such as when machining high-hardness alloys. Increase, make More accurately reflects cutting load; through The system's online learning mechanism allows it to automatically adapt to different materials, such as aluminum, steel, and titanium alloys, eliminating the need for preset parameters while maintaining computational simplicity.

[0060] like Figure 3 As shown, the horizontal axis represents time, and the vertical axis represents the state response intensity. During the waiting or no-load stage, the state response intensity approaches 0, indicating that the interference of basic energy consumption and material hardness has been effectively eliminated. During the cutting stage, the state response intensity shows a stable positive value, and the fluctuation range remains consistent under different material hardnesses, proving that the material hardness adaptive compensation mechanism effectively eliminates the influence of material characteristic differences. During the tool wear warning stage, the state response intensity shows a continuous fluctuation that deviates from the normal range, which directly reflects the abnormality of production actions and verifies its ability to accurately capture the essential characteristics of the production state.

[0061] S004: Based on a sliding window dynamically determined by the signal spectrum, the state response intensity is normalized to generate normalized state features.

[0062] It should be noted that the state response intensity may still include environmental noise, such as workshop vibration, sensor electromagnetic interference, and transient interference during operating condition switching, such as instantaneous fluctuations caused by sudden changes in speed. These non-production-related disturbances can disrupt the continuity of the state temporal pattern, leading to misjudgments by the LSTM model. Therefore, this step requires normalization processing, combined with dynamic windows and adaptive noise tolerance, to suppress interference signals, highlight the stable temporal characteristics of the production state, and improve the robustness of model recognition.

[0063] Specifically, the normalized state characteristics satisfy:

[0064] ;

[0065] In the formula, For the current moment The normalized state characteristics, For the current moment State response intensity, For a moment State response intensity, The length of the sliding window is dynamically determined based on the signal spectrum. The state response intensity signal is analyzed by short-time Fourier transform (STFT), and the dominant frequency is extracted. ,calculate ; For the current moment The mean of the state response intensity at all times within the sliding window. The noise tolerance is set.

[0066] Specifically, the noise tolerance is obtained as follows:

[0067] Under no-load conditions on the production line, the state response intensity is collected for a corresponding duration.

[0068] Calculate the standard deviation of the state response intensity under the no-load condition;

[0069] The result of 1.5 times the standard deviation is used as the noise tolerance.

[0070] in, The numerical characteristics are highly correlated with the production state. When the equipment is in a stable production state, such as finishing or uniform speed cutting, The signal fluctuations are gentle, and the variance within the sliding window is small. It exhibits stable high values, making it easy for LSTM to capture typical timing patterns of this production state; when the equipment is undergoing operating condition switching, such as speed adjustment or strong noise interference scenarios, The fluctuations are dramatic, and the variance within the sliding window increases. The values ​​are suppressed to low values, effectively preventing LSTM from misinterpreting non-business-related fluctuations as state changes.

[0071] like Figure 4 As shown, the vertical axis represents the normalized state features. During the steady-state cutting stage, the normalized state features exhibit stable high values, indicating that the dynamic sliding window adapts to the spectral characteristics of the cutting signal, significantly amplifying the steady-state production action features. During the working condition switching or transient interference stage, the normalized state features rapidly decrease to low values, effectively suppressing non-production-related transient fluctuations and preventing interference signals from affecting model recognition. This verifies the effectiveness of the dynamic sliding window and noise tolerance design, ensuring stable feature distribution and adapting to the switching requirements of multi-variety, small-batch production.

[0072] S005: Input the normalized state features into the pre-trained long short-term memory network model to identify the current production status of the production line and complete the real-time monitoring of the production status based on the aggregation of production line data.

[0073] It should be noted that, since the normalized state features have eliminated the influence of operating condition variables such as rotational speed and material, the LSTM model no longer needs to learn currents greater than [a certain value]. It is not coarse processing, which depends on the rules of working conditions, but learning the energy deviation to present a specific trapezoidal waveform is a generalized temporal pattern of coarse processing. The LSTM model extracts the evolution law of normalized state features in the time dimension through its forget gate and input gate mechanism, and outputs the production logic state probability vector corresponding to the current moment.

[0074] Specifically, identifying the current production status of the production line includes:

[0075] Normalized state features from multiple consecutive time steps, including the current time step, are input into a Long Short-Term Memory (LSTM) network model. The probability vector output by the LSM model is then processed. Normalization processing;

[0076] The state with the highest probability is selected as the current preliminary prediction result. To prevent jumps caused by instantaneous interference, a majority voting mechanism is used to evaluate the continuous probability. The prediction results at each time point are smoothed to finally output a definite production status, such as normal processing, standby, fault alarm, etc., thus completing real-time monitoring of production status based on production line data aggregation.

[0077] This invention also discloses a real-time production status monitoring system based on production line data aggregation, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a real-time production status monitoring method based on production line data aggregation according to this invention is implemented.

[0078] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0079] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of production status based on production line data aggregation, characterized in that, include: Real-time acquisition of production line data including spindle current, spindle command speed, and workpiece material hardness coefficient; Based on pre-calibrated equipment dynamic parameters, a theoretical energy consumption benchmark reflecting the basic energy consumption under the current operating conditions of the equipment is constructed using the spindle command speed. The theoretical energy consumption benchmark is composed of a speed-related steady-state energy consumption component and an acceleration-related dynamic energy consumption component, satisfying the following: ； For the current moment The theoretical energy consumption benchmark, This refers to the steady-state energy consumption component related to rotational speed. For acceleration-related dynamic energy dissipation components, For the current moment spindle command speed, For the current moment The difference between the spindle speed commanded at the previous moment and the speed commanded at the previous moment. The sampling interval is... The frictional resistance correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is... The moment of inertia correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the relationship between acceleration and energy consumption; The spindle current is differentially processed with the theoretical energy consumption benchmark, and adaptive compensation is performed in combination with the workpiece material hardness coefficient to calculate the state response intensity. The state response intensity characterizes the business action characteristics after the peeling condition and material property interference. The state response strength satisfies: ； For the current moment State response intensity, For the current moment spindle current, The rated line voltage on the input side of the spindle drive. The current moment is dynamically determined through online learning. The influence coefficient of material hardness, For the current moment The corresponding workpiece material hardness coefficient, The hyperparameter constant is used to prevent the denominator from being zero. It is the natural logarithm function; The hardness influence coefficient is a dynamically adjustable parameter, and the adjustment method is as follows: When a cutting state is detected, recent data is collected. The state response intensity at each cutting moment and the corresponding workpiece material hardness coefficient; calculate The average ratio of the state response intensity to the corresponding workpiece material hardness coefficient at each cutting moment is used as the material hardness influence coefficient at the current moment. Based on a sliding window dynamically determined by the signal spectrum, the state response intensity is normalized to generate normalized state features. Normalized state features are input into a pre-trained long short-term memory network model to identify the current production status of the production line and complete real-time monitoring of the production status based on the aggregation of production line data.

2. The method for real-time monitoring of production status based on production line data aggregation according to claim 1, characterized in that, The calibration method for the device's dynamic parameters is as follows: When the production line is running under no-load conditions, the spindle current and spindle command speed are collected synchronously. Based on the collected data, the least squares method is used to fit the equation. Thus determine and The value; In the formula, Main spindle current, Main spindle speed command This is the difference between the spindle command speed at the sampling time and the previous sampling time, used to calibrate the equipment's dynamic parameters. The sampling interval is... The frictional resistance correlation coefficient, which is used in pre-calibrated equipment dynamic parameters to reflect the contribution of the square of rotational speed to energy consumption, is... The moment of inertia correlation coefficient, which is a pre-calibrated dynamic parameter of the equipment and reflects the relationship between acceleration and energy consumption.

3. The method for real-time monitoring of production status based on production line data aggregation according to claim 1, characterized in that, It also includes initialization and maintenance processing for the hardness influence coefficient: During system initialization, a fixed initial value is assigned to the hardness influence coefficient from the material database; When the equipment is in a non-cutting state, the hardness influence coefficient is updated using an exponential decay method until it reaches the set lower limit value.

4. The method for real-time monitoring of production status based on production line data aggregation according to claim 1, characterized in that, The normalized state characteristics satisfy: ； In the formula, For the current moment The normalized state characteristics, For the current moment State response intensity, For a moment State response intensity, The length of the sliding window. For the current moment The mean of the state response intensity at all times within the sliding window. The noise tolerance is set.

5. The method for real-time monitoring of production status based on production line data aggregation according to claim 4, characterized in that, The noise tolerance is obtained as follows: Under no-load conditions on the production line, the state response intensity is collected for a corresponding duration. Calculate the standard deviation of the state response intensity under the no-load condition; The result of 1.5 times the standard deviation is used as the noise tolerance.

6. The method for real-time monitoring of production status based on production line data aggregation according to claim 1, characterized in that, The method of identifying the current production status of the production line includes: Normalized state features from multiple consecutive time steps, including the current time step, are input into a Long Short-Term Memory (LSTM) network model. The probability vector output by the LSM model is then processed. Normalization processing; A majority voting mechanism is used to smooth the prediction results over multiple consecutive time points to determine the current production status of the production line.

7. A real-time production status monitoring system based on production line data aggregation, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions, which, when executed by the processor, implement a real-time production status monitoring method based on production line data aggregation according to any one of claims 1-6.

Images