A digital-based intelligent management and control method and system for artillery production workshops

Abstract

This invention discloses a digitally-based intelligent control method and system for artillery production workshops, relating to the field of workshop management technology. The method includes collecting multi-source operational status data and key quality inspection parameter data from processing equipment and workstations in the artillery production workshop, performing time-aligned processing to form a unified time-series workshop status data set. Based on this data set, an equipment health index model is constructed using adaptive fusion of digital twin residuals and multi-dimensional features to obtain an equipment health vector. Furthermore, based on recursive filtering and virtual extrapolation of future operating conditions, the remaining lifespan vector is obtained. The cycle time deviation is calculated based on the actual processing time and the preset standard cycle time, and the equipment load rate is calculated by combining the task arrival rate and equipment capacity parameters. This invention significantly improves the accuracy and robustness of artillery workshop control through multi-dimensional status perception and dynamic collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of workshop management technology, and in particular to a digitally based intelligent control method and system for artillery production workshops. Background Technology

[0002] With the deep integration of informatization and industrialization, digital manufacturing has become a core means to improve the production capacity of complex equipment. Artillery, as a typical complex electromechanical product, involves multiple stages in its production workshop, including precision machining, assembly, and testing, characterized by tightly coupled processes, dense equipment, and extremely high requirements for quality consistency. In recent years, technologies such as digital twins, manufacturing execution systems, and equipment condition monitoring have been gradually applied in artillery manufacturing. By deploying sensors on key equipment, real-time acquisition and visual monitoring of operating parameters such as vibration, temperature, and current have been initially achieved. Some advanced workshops are also attempting to introduce equipment health management concepts, using statistical feature analysis to conduct post-assessment of equipment degradation trends.

[0003] However, current technological development remains primarily focused on "state awareness" and "data recording," namely, achieving transparency in the production process through Manufacturing Execution Systems (MES) and fault alarms and maintenance records through Equipment Management Systems (EMS). In practical applications, significant functional limitations and technical bottlenecks persist. Firstly, in equipment health assessment, traditional methods mostly rely on raw characteristics or simple statistics such as vibration amplitude and temperature, failing to adequately consider the interference of operating conditions on these characteristic values ​​and thus failing to accurately reflect the actual degradation trend of the equipment. Secondly, in remaining life prediction, existing technologies often employ fixed-window differencing to estimate degradation rates or extrapolate from black-box models trained on historical data. The former is extremely sensitive to measurement noise, and short-term fluctuations easily lead to distorted rate estimates; the latter cannot predict the impact of future production plans on the equipment degradation process, inevitably resulting in significant deviations between life predictions and actual conditions, affecting the scientific validity of preventative maintenance and control decisions. Therefore, the problem of distorted health assessments and life predictions caused by operating condition fluctuations remains to be addressed. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a digital-based intelligent management and control method and system for artillery production workshops, which solves the problems of distorted health assessment and deviation in life prediction caused by fluctuations in operating conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a digitally based intelligent management and control method for artillery production workshops, comprising,

[0008] Multi-source operational status data and key quality inspection parameter data of processing equipment and workstations in the artillery production workshop are collected and time-aligned to form a unified time-series workshop status data set. Based on the workshop status data set, an equipment health index model based on digital twin residual and multi-dimensional feature adaptive fusion is constructed to obtain the equipment health vector. Based on recursive filtering and virtual extrapolation of future working conditions, the remaining life of the equipment is predicted to obtain the remaining life vector.

[0009] The cycle time deviation is calculated based on the actual processing time and the preset standard cycle time. The equipment load rate is calculated by combining the task arrival rate and equipment capacity parameters. A set of production execution deviations is constructed by the equipment remaining life vector, cycle time deviation, and equipment load rate.

[0010] The quality consistency index is calculated based on key quality inspection parameter data as a quality stability indicator, thereby comprehensively quantifying quality fluctuations and mean deviations.

[0011] Based on the set of production execution deviations and quality stability indicators, a control objective function is constructed, and the task allocation and processing parameters are optimized and solved to generate a control vector.

[0012] The control vector is converted into equipment control commands and issued for execution. The model parameters and control thresholds are updated based on the execution feedback to form a continuous closed-loop control process.

[0013] As a preferred embodiment of the digital-based intelligent control method for artillery production workshops described in this invention, the multi-source operating status data includes spindle speed, spindle current, vibration acceleration, bearing temperature, and actual processing time of a single piece, while the key quality detection parameter data is the average value of the key quality parameters output by the detection station.

[0014] The collected data is time-aligned to form a unified time-series set of workshop status data.

[0015] As a preferred embodiment of the digital-based intelligent management and control method for artillery production workshops described in this invention, the steps for forming equipment health vectors and lifespan vectors are as follows:

[0016] Based on the workshop condition data set, vibration energy characteristics, impact characteristics, temperature rise trend characteristics, and absolute temperature characteristics were extracted and normalized.

[0017] Based on normalized features, a digital twin reduced-order model is constructed to generate theoretical normal feature benchmark values ​​and calculate feature residuals.

[0018] Based on feature residuals, a lightweight attention mechanism based on information entropy is used to adaptively update the weights and calculate the health index to obtain the device health vector.

[0019] Based on the equipment health vector, the remaining lifetime vector is obtained through recursive filtering and virtual extrapolation of future operating conditions to predict the remaining lifetime.

[0020] As a preferred embodiment of the intelligent control method for artillery production workshop based on digitalization described in this invention, the cycle deviation refers to the deviation calculated based on the actual cycle and the standard cycle, and processes exceeding the threshold are marked as cycle deviation exceeding the limit.

[0021] As a preferred embodiment of the intelligent management and control method for artillery production workshop based on digitalization described in this invention, the generation of the production execution deviation set refers to the statistical calculation of equipment load based on tasks and capabilities, the marking of status according to thresholds, and the summarization of cycle deviation, equipment load, and remaining usable life of the equipment into the production execution deviation set.

[0022] As a preferred embodiment of the intelligent control method for digital artillery production workshop described in this invention, the method of calculating the quality consistency index based on key quality inspection parameter data as a quality stability indicator refers to calculating the standard deviation and the quality consistency index, and triggering a quality drift warning signal when the threshold is exceeded.

[0023] As a preferred embodiment of the intelligent control method for artillery production workshop based on digitalization described in this invention, the step of constructing a control objective function based on the production execution deviation set and quality stability index refers to normalizing the production execution deviation set and quality stability index, and constructing the control objective function by weighted summation.

[0024] As a preferred embodiment of the intelligent control method for artillery production workshop based on digitalization described in this invention, the step of optimizing and solving the task allocation and processing parameters to generate the control vector refers to generating the control vector using a hierarchical deterministic process of "first task allocation, then parameter fine-tuning".

[0025] As a preferred embodiment of the intelligent control method for artillery production workshop based on digitalization described in this invention, wherein: the formation of continuously iterating closed-loop control refers to converting control vectors into equipment control commands and issuing them for execution;

[0026] Based on the execution confirmation status and the updated field data, the cycle time deviation, equipment load rate, and quality consistency index are recalculated to form a continuously iterative closed-loop control.

[0027] Secondly, this invention provides a digital-based intelligent control system for artillery production workshops, comprising:

[0028] The data acquisition and alignment module is used to collect multi-source operating status and quality data, perform time alignment processing, and form a unified time-series workshop status data set;

[0029] The health life assessment module is used to calculate cycle time deviation and equipment load rate, and combined with remaining life constraints, to generate a set of production execution deviations containing multi-dimensional information;

[0030] The quality stability quantification module is used to calculate the quality consistency index based on key quality inspection parameters, comprehensively quantify quality fluctuations and mean deviations, and generate quality stability indicators.

[0031] The closed-loop optimization control module is used to construct the control objective function for optimization, generate control vectors and issue them for execution, and update parameters based on feedback to form a continuous closed loop.

[0032] The beneficial effects of this invention are as follows: By constructing a digital twin residual health index, this invention removes operating condition interference and achieves accurate perception of equipment degradation trends; by integrating future production plans to extrapolate remaining lifespan, it deeply couples the prediction results with real-time scheduling scenarios; ultimately, it can provide early warning of equipment health risks and dynamically adjust processing load and maintenance windows, transforming passive response into proactive control, thereby ensuring high reliability and continuity of the entire artillery production workshop process. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of a digitally based intelligent control method for an artillery production workshop, as shown in Example 1.

[0035] Figure 2 This is a structural diagram of a digital-based intelligent control system for an artillery production workshop, as shown in Example 1.

[0036] Figure 3 This is the process for constructing the equipment health index and predicting remaining lifespan in Example 1. Detailed Implementation

[0037] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0038] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0040] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides a digitally based intelligent management and control method for artillery production workshops, including the following steps:

[0041] S1. Collect multi-source operating status data and key quality inspection parameter data of processing equipment and workstations in the artillery production workshop and perform time alignment processing to form a unified time series of workshop status data set. Based on the workshop status data set, construct an equipment health index model based on digital twin residual and multi-dimensional feature adaptive fusion to obtain the equipment health vector. Based on recursive filtering and future working condition virtual extrapolation, predict the remaining life of the equipment to obtain the remaining life vector.

[0042] S1.1: Multi-source operating status data includes spindle speed n(t), spindle current I(t), vibration acceleration a(t), and bearing temperature. Actual processing time per piece The key quality inspection parameter data is the mean value of the key quality parameters output by the inspection station, Mean(t).

[0043] Specifically, in each fixed sampling period Industrial data acquisition terminals are installed at key processing equipment (such as deep hole machining centers), assembly stations, and testing stations within the artillery production workshop to collect multi-source operational status data and key quality inspection parameter data in real time. The key quality inspection parameters are quality inspection values ​​used to characterize the accuracy and consistency of artillery products. At least one key quality characteristic is preferentially selected as the inspection object, such as one or more of the following: barrel caliber, coaxiality, roundness, or surface roughness. The testing station can output a set of multiple measurements of the key quality characteristic within the current sampling period. To facilitate subsequent consistency calculations, the arithmetic mean of multiple measurement sets within the sampling period is taken to obtain the mean value Mean(t).

[0044]

[0045] In the formula, is the value of the r-th detection, and M is the number of detections within the sampling period.

[0046] It should be noted that, This system is designed based on a comprehensive engineering assessment of the dynamic response characteristics of artillery processing equipment, the load capacity of industrial network communication, and the real-time requirements of quality control. Following the principle of "balancing fault detection sensitivity and system resource efficiency," theoretical analysis is compared with actual field measurement data to obtain the optimal value range. The specific values ​​within this range are then fine-tuned according to the actual operating conditions of the production line and solidified into the system. For example, in a conventional artillery machining production line, the following values ​​can be set... In high-dynamic or critical process segments, the encryption can be temporarily reduced to 0.5s, reflecting the principle of "normal steady-state sampling and abnormal transient encryption".

[0047] S1.2: Perform time alignment processing on the collected data to form a unified time-series set of workshop status data.

[0048] Specifically, at the end of each sampling period, the dimensionless mutation index S(t) is calculated. When the dimensionless mutation index S(t) is greater than or equal to the mutation threshold, the mutation is considered complete. If a rapid change trend is detected, the timer is reset; if the timer expires and is not triggered again, it automatically reverts to a fixed sampling period. This improves sampling resolution when the device status changes rapidly. It adds a unified timestamp generated by the acquisition terminal to each piece of acquired data and performs alignment processing.

[0049] It should be noted that the alignment rule is as follows: each data entry is mapped to its nearest-time raster. When multiple data points appear within the same grid, the arithmetic mean of the data points in that grid is taken as the grid's representative value. When a grid is missing a measurement, the representative value of the previous grid is used to maintain continuity (zero-order maintenance). This time grid alignment method is a commonly used data synchronization technique in industrial settings. It is computationally simple and easy to implement in engineering, avoiding misjudgments caused by "time misalignment" between different data sources in subsequent modeling. After data acquisition, mutation detection, and time alignment processing, a unified time-series set of workshop status data is obtained. .

[0050] The expression for calculating the dimensionless mutation index is:

[0051]

[0052] In the formula, S(t) is a dimensionless mutation index. and These are the vibration accelerations for the current and previous sampling periods, respectively. This is the equipment vibration reference value. and These are the bearing temperatures for the current and previous sampling periods, respectively. This is a temperature reference value.

[0053] It should be noted that, The 95th percentile of the statistical distribution of S(t) during the historical stable operation period of the equipment is taken as the initial threshold. The typical value range is [0.05, 0.15]. Values ​​below 0.05 are easily affected by high-frequency noise interference, which may lead to false triggering. Values ​​above 0.15 are not sensitive to early weak impacts. The preferred value is 0.08.

[0054] Taking into account the matching relationship between the industrial Ethernet clock synchronization accuracy (typically <10ms) and the encryption sampling period (500ms), the typical value range is [50ms, 500ms]. Too small a value is limited by network jitter, while too large a value loses the significance of multi-source data synchronization and alignment. The preferred value is 100ms.

[0055] Take the 95th percentile of the equipment's rated vibration limit or historical stable operating value.

[0056] Use the highest permissible operating temperature of the equipment.

[0057] S1.3: Based on the workshop status data set, extract vibration energy characteristics, impact characteristics, temperature rise trend characteristics, and absolute temperature characteristics and perform normalization processing.

[0058] It should be noted that this step aims to reduce the interference of fluctuations in operating conditions such as speed and load on health assessment and to achieve second-level adaptive updates of weights, so that the health index mainly reflects the actual degradation trend of the equipment and does not reflect changes in normal operating conditions. The digital twin residual refers to the deviation between the "measured degradation characteristic value" and the "theoretical normal characteristic benchmark value" under the current operating conditions. The adaptive fusion refers to automatically adjusting the fusion weights based on the changes in the amount of information in each characteristic residual within a short time window.

[0059] Specifically, based on workshop status data sets For each device, a multidimensional degradation feature vector is extracted within the sliding time window L, and the degradation features are subjected to min-max normalization to eliminate the influence of dimensions.

[0060] It should be noted that the multidimensional degradation feature vector includes vibration energy features, impact features, temperature rise trend features, and absolute temperature features;

[0061] The vibration energy characteristic is expressed using the root mean square value of vibration, and the calculation formula is as follows:

[0062]

[0063] In the formula, It is a characteristic of vibrational energy. Is the i-th device at time... The vibration acceleration is K, where K is the number of sampling points within the window.

[0064] Impact characteristics are determined using vibration kurtosis, calculated using the following formula:

[0065]

[0066] In the formula, It is an impact characteristic. It is the mean vibration value within the window. It is the standard deviation of vibration within the window.

[0067] The temperature rise trend is characterized by the rate of temperature change, calculated using the following formula:

[0068]

[0069] In the formula, It is an absolute temperature characteristic. This is the sampling time interval; multiply by 3600 for unit conversion.

[0070] The absolute temperature characteristic is based on the average bearing temperature, and the calculation formula is as follows:

[0071]

[0072] In the formula, It is an absolute temperature characteristic;

[0073] The minimum-maximum normalization process is as follows:

[0074]

[0075] In the formula, Normalization characteristics , The data are taken from the historical statistical extreme values ​​of the equipment when it had passed accuracy calibration and was in a stable production stage.

[0076] It should be noted that the stable production stage refers to the period during which the equipment is free of known faults, has stable quality, and operates normally. Selecting the statistical extreme value of this stage as the normalization benchmark can make the characteristic scale consistent with the operating conditions of the equipment itself, and avoid the scale differences between different equipment from affecting healthy integration.

[0077] S1.4: Based on normalized features, construct a digital twin reduced-order model to generate theoretical normal feature benchmark values ​​and calculate feature residuals;

[0078] It should be noted that a polynomial regression model that is interpretable and easy to implement in engineering is preferred. With rotational speed and current as inputs, a theoretical baseline function is established for each normalized feature. After obtaining the theoretical baseline function, the residual is generated as the actual degradation input.

[0079] The theoretical benchmark function is:

[0080]

[0081]

[0082] In the formula, It is the theoretical normal reference value of the e-th normalized feature under the current operating conditions. These are the regression coefficients, n(t) is the rotational speed, and I(t) is the current.

[0083] The actual degenerate input is:

[0084]

[0085] In the formula, It is the residual of the e-th normalized feature. The larger the residual, the more the feature deviates from the theoretical normal trajectory and the higher the risk of degradation.

[0086] It should be noted that the regression coefficients are obtained through offline training using the least squares method on a stable running sample set to minimize the mean square error. Through offline training and online calling, a corresponding dynamic benchmark can be output for each working condition without increasing the on-site computational burden.

[0087] To minimize the mean square error, the following is required:

[0088]

[0089] In the formula, It is a set of stable running samples.

[0090] Use residuals instead of directly using normalized features By integrating these factors, "normal increases caused by working conditions" can be removed from health assessments, making the health index more sensitive to actual deterioration.

[0091] S1.5: Based on feature residuals, a lightweight attention mechanism based on information entropy is used to adaptively update the weights and calculate the health index to obtain the device health vector;

[0092] Specifically, the nearest O residual points are taken to form the entropy calculation window. The residual values ​​are divided into B bins of equal width. The bin probability is calculated, and the Shannon information entropy is calculated. The dynamic weights are obtained through Softmax normalization.

[0093] The binning probability is:

[0094]

[0095] In the formula, It is the binning probability of the e-th normalized feature residual in the d-th bin. It is the frequency of the landing point of the e-th normalized feature residual in the d-th bin.

[0096] The Shannon information entropy is:

[0097]

[0098] In the formula, It is the information entropy of the e-th normalized feature residual.

[0099] The dynamic weight is:

[0100]

[0101] In the formula, It is the dynamic weight of the e-th normalized feature residual. and This enables real-time adaptive fusion of multi-feature residuals.

[0102] It should be noted that when a certain sub-box appears When the information entropy is zero, the higher the information entropy, the higher the disorder and information content of the residual sequence within the short window. This usually corresponds to more significant abnormal fluctuations or degradation trends, and therefore should be given a higher fusion weight.

[0103] Furthermore, considering the "thermal-vibration co-deterioration" effect, and combining nonlinear coupling terms to enhance sensitivity to high-thermal-vibration conditions, an equipment health index is constructed, forming an equipment health vector. .

[0104] The health index is defined as:

[0105]

[0106] In the formula, It is the health index of the i-th device at time t. It is the residual vibrational energy. It is the temperature mean residual. It is the coupling strength coefficient. The calculation results are limited to the range of [0,1] to prevent the health index from exceeding the limit due to instantaneous shocks.

[0107] It should be noted that, A deterministic offline calibration method is used: on a historical dataset containing stable and degenerate segments, let A grid search is performed within the range [0.1, 1.0] with a step size of 0.1, and different... The monotonicity index of the health index series and the life prediction error are combined, and the optimal value is taken as the final value, which is 0.5.

[0108] Defined as:

[0109]

[0110] In the formula, · represents the input value.

[0111] S1.6: Based on the equipment health vector, the remaining lifetime vector is obtained by recursive filtering and virtual extrapolation of future operating conditions to predict the remaining lifetime.

[0112] It should be noted that this step aims to address the issues of the fixed window difference method being sensitive to noise and the inability of the simple extrapolation method to predict future load changes; by smoothing the degradation rate estimate through recursive filtering and combining it with the load spectrum of future production plans for discrete step extrapolation, a remaining lifetime more consistent with the actual scheduling is obtained.

[0113] Specifically, after obtaining a continuous health index sequence, an exponential smoothing recursive filter with a forgetting factor is used to estimate the degradation rate. After obtaining the degradation rate, the production scheduling plan for the future time window is read from the MES system, and the expected rotational speed and load sequence are extracted to form the future load spectrum. The current health index is used as the initial value for the extrapolation. In each extrapolation step, a nonlinear degradation recursive equation is used to update the health index, with a failure threshold set. When the updated health index is greater than or equal to the failure threshold for the first time during the simulation process... If the time is right, record that moment and calculate the remaining lifetime to form the equipment lifetime vector. To ensure that lifetime estimates can still be output even in scenarios with no future plans, if the future load spectrum cannot be obtained, an analytical approximation method is used to calculate the remaining lifetime.

[0114] The degradation rate is:

[0115]

[0116] In the formula, It is the rate of decline in health indicators. It is the rate of decline in health index from the previous cycle. It is a forgetting factor. It controls the time interval of the cycle.

[0117] The future load spectrum is .

[0118] The updated health index is:

[0119]

[0120] In the formula, It is an updated health index. It is the deduction step size. is the degradation acceleration coefficient, and m is the degradation acceleration exponent.

[0121] The remaining lifetime is:

[0122]

[0123] In the formula, It is the remaining lifespan. It is the moment of failure.

[0124] The analytical approximation is:

[0125]

[0126] It should be noted that when the rate of decline in health index... When the health index does not rise or fluctuates and declines, in order to avoid divergence in life expectancy estimates, the rate of decline in the health index is... Set to a minimum positive value To maintain computability.

[0127] The analytical expression is a first-order approximation of the discrete derivation recursion under small step conditions, which can provide a stable and computable baseline result in the absence of a plan.

[0128] Based on the protective settings of numerical calculation stability and natural aging assumptions, the typical value range is: The preferred value is 0.001 / day. This is based on the assumption of natural aging to prevent the denominator from being zero. It ensures that when the equipment is healthy or its condition improves, the system can still output a finite and reasonable "ultra-long lifespan" estimate, thus guaranteeing the numerical stability of the algorithm.

[0129] Based on the normalization setting of the model's domain and physical failure boundary, the typical value range is [0.9, 1.0]. This is because, in the normalized model, 1.0 represents the theoretical absolute limit. To leave a certain safety margin, it can sometimes be set to 0.95, but under the clip mechanism of this model, 1.0 is the cutoff point, representing "complete failure," and the preferred value is 1.0.

[0130] The fitting setting of m is based on the nonlinear characteristics of the PF curve (potential fault-functional fault curve). The typical value range is [0.5, 2.0], used to adjust the initial influence of the nonlinear term, with a preferred value of 1.0. The typical value range of m is [1.0, 3.0], where 1.0 represents linear degradation, applicable only to a very few cases; 2.0 represents quadratic degradation, consistent with the fatigue spalling law of most rolling bearings; and 3.0 represents cubic degradation, suitable for conditions with extremely high impact loads, with a preferred value of 2.0.

[0131] Furthermore, to adapt to the time-varying degradation patterns, residual consistency detection and online parameter correction are used, employing the degradation rate of the health index from the previous period. With the present The health index of the previous period is recursively predicted one step by m, the residual is calculated, the residual variance is calculated within the sliding window, and compared with the benchmark variance to trigger parameter updates.

[0132] The step-by-step recursive prediction of the health index for the previous period is as follows:

[0133]

[0134] In the formula, It is a theoretical value of the predicted health index. It is the health index raised to the power of m at the previous moment.

[0135] The residual is:

[0136]

[0137] In the formula, It is a residual.

[0138] The trigger condition for parameter update is:

[0139]

[0140] In the formula, It is the residual variance. If the residual variance is the baseline during the stable operation phase, then it is determined that the degradation mode has drifted and a parameter update is triggered. It is the variance amplification factor threshold.

[0141] The parameters are updated as follows:

[0142]

[0143] In the formula, It is the updated set of parameters for the degradation model (including parameters related to the equivalent degradation rate). These are parameter values ​​estimated using the least squares method based on the most recent window data. It is a parameter smoothing coefficient used to ensure the continuity of parameter updates.

[0144] It should be noted that, It is obtained by estimating the equivalent degradation parameters in the recursive equation using the least squares method within the nearest window. The updated degradation rate level is used to check and correct the degradation rate of the health index in the next cycle, so that the life prediction remains accurate in the long term.

[0145] Using "Statistical Process Control (SPC)" The "principle and hypothesis testing" technique is set with a typical value range of [1.5, 4.0]. If the threshold is too low, it is easily affected by high-frequency noise, which may cause the model to update frequently and oscillate. If the threshold is too high, it will be slow to respond to sudden changes in the degradation mode, resulting in a lag in correction. The preferred value is 2.25 (corresponding to 1.5 times the standard deviation limit), which can effectively filter out regular fluctuations while ensuring timely correction when the error distribution is significantly divergent.

[0146] The filter is set based on the principle of exponentially weighted moving average (EWMA), with a typical value range of [0.6, 0.95]. If the weight of the new estimate is too large, it is easily affected by short-term noise in the least squares estimation, resulting in drastic parameter fluctuations (overfitting). If the weight of the historical parameters is too large, the model will track new degradation patterns too slowly (underfitting). The optimal value is 0.8, which retains 80% of the historical information to maintain stability, while introducing 20% ​​of the new information for gradual correction. This is the best engineering compromise for achieving online self-evolution of the model without divergence.

[0147] S2. Calculate the cycle time deviation based on the actual processing time and the preset standard cycle time, and calculate the equipment load rate by combining the task arrival rate and equipment capacity parameters. Construct a production execution deviation set through the equipment remaining life vector, cycle time deviation, and equipment load rate.

[0148] S2.1: Based on the deviation calculated between the actual and standard cycle time, processes exceeding the threshold are marked as cycle time deviation exceeding the limit.

[0149] Specifically, for each process s (or the processing task completed by the corresponding equipment) in the current statistical period Within, based on the actual processing time of a single piece after collection and alignment. And read the standard cycle time of the process from the process document or the pre-set standard process cycle time table. Calculate the beat deviation When the cycle time deviation is greater than 0, it indicates that the actual time taken for this process exceeds the standard cycle time, and there is a tendency for delay. To facilitate subsequent control and judgment, an allowable deviation threshold is set. ,when When this happens, the process is marked as "cycle deviation exceeds the limit".

[0150] The formula for calculating beat deviation is:

[0151]

[0152] In the formula, It is the cycle time deviation of the s-th process.

[0153] Allowable deviation threshold The calculation formula is:

[0154]

[0155] In the formula, U represents the urgency of the order. It is the basic allowable deviation coefficient. It is the urgency gain coefficient.

[0156] The formula for calculating the urgency level U of an order is:

[0157]

[0158] In the formula, It's the order's required delivery time. It is the current time. It is the standard production cycle of an order, which is obtained by adding up the standard cycle time of the process route. This limits the result to between 0 and 1.

[0159] It should be noted that, The "production cycle matching and statistical significance trade-off method" is used for setting, with a typical value range of [30min, 120min], and the preferred value is 60min, which is the best balance point between 'data stability' and 'real-time control'.

[0160] The "process capability index (Cpk) statistical back-calculation method" was used for setting, with a typical value range of [0.03, 0.08] and a preferred value of 0.05. Based on measured data from a certain type of artillery production line, the coefficient of variation (CV) of processing time under normal operating conditions is approximately 2.0%~2.5%. The principle (covering 95% of normal fluctuations) was ultimately selected.

[0161] The setting adopts the "production scheduling flexibility and delivery risk trade-off method," with a typical value range of [0.05, 0.15], and a preferred value of 0.10. 0.15 is the recognized "limit tolerance boundary" in artillery manufacturing technology—exceeding this limit usually means a serious imbalance in process parameters or uncontrollable quality risks. Setting it to 0.10 can provide sufficient continuous production buffer space for rush orders (avoiding frequent shutdowns due to minor fluctuations) while strictly limiting the maximum deviation within a safe range, reflecting a dynamic balance between "ensuring delivery" and "ensuring quality."

[0162] S2.2: Statistical task and capacity calculation of equipment load, marking status according to threshold, and summarizing cycle deviation, equipment load and remaining usable life of equipment into a production execution deviation set;

[0163] Specifically, for each device, in the current statistical period Internal statistics on the number of tasks entering the device queue or assigned to the device. Calculate the task arrival rate; simultaneously, obtain the equipment's processing capacity per unit time under the corresponding processing mode from the equipment capacity parameter table or historical stable operation data. The equipment load rate was calculated; to ensure the stability of the artillery manufacturing process and to reserve a disturbance buffer, a safe operating limit was set. When the equipment load rate is less than the safe operating limit When the equipment load rate exceeds the safe operating limit, the equipment is in a safe operating zone. When the load rate is less than 1, it is marked as a load warning. When the equipment load rate is greater than or equal to 1, the equipment is marked as a "bottleneck device," and the cycle time deviation is recorded. Equipment load rate and the remaining usable life of the equipment This is used to compile a set of production execution deviations. .

[0164] The formula for calculating the mission arrival rate is:

[0165]

[0166] In the formula, It is the task arrival rate.

[0167] The formula for calculating equipment load rate is:

[0168]

[0169] In the formula, It refers to the equipment load rate.

[0170] It should be noted that, The "nonlinear inflection point analysis method for queuing time" is adopted for setting, with a typical value range of [0.80, 0.90] and a preferred value of 0.85. When the load rate exceeds 0.85, the average waiting time increases exponentially, and the system stability drops sharply. Reserving 15% of the production capacity redundancy as a buffer can effectively absorb random disturbances such as tool changes, material fluctuations, and emergency orders in artillery production, and prevent the backlog of work-in-process.

[0171] S3. Calculate the quality consistency index based on key quality inspection parameter data as a quality stability indicator, thereby comprehensively quantifying quality fluctuations and mean deviations.

[0172] S3.1: Calculate the standard deviation and quality consistency index, and trigger a quality drift warning signal if the threshold is exceeded.

[0173] Specifically, based on a set of multiple measurements The standard deviation is calculated from the mean (Mean(t). When the number of tests M=1 within the sampling period, the standard deviation is set to 0 to ensure that the calculation is feasible and does not introduce meaningless fluctuations. To determine whether the current fluctuation is abnormal, historical data of stable equipment and process operation are selected, the standard deviations are calculated separately, and the arithmetic mean is taken as the steady-state baseline standard deviation. To simultaneously reflect both "dispersion (fluctuation)" and "center shift (mean shift)," a quality consistency index is calculated, and a quality consistency threshold is set. When the quality consistency index is greater than the quality consistency threshold When this happens, a mass drift warning signal is triggered. .

[0174] The formula for calculating standard deviation is:

[0175]

[0176] In the formula, It is the standard deviation;

[0177] The formula for calculating the steady-state reference standard deviation is:

[0178]

[0179] In the formula, It is the standard deviation of the l-th historical window. It refers to the number of windows;

[0180] The formula for calculating the quality consistency index is:

[0181]

[0182] In the formula, CI(t) is the quality consistency index. It is the normalized amount of mean shift, exp( ) is an exponential function, Target is the mean offset penalty coefficient, Tol is the target value of the key quality characteristic read from the product drawing / process document, and Tol is the tolerance half-width of the characteristic.

[0183] It should be noted that the quality consistency index CI(t) serves as a "quality stability index" characterizing the processing stability of the current process. This index, through a product coupling method, incorporates the "quality fluctuation" term, which reflects the degree of dispersion. The "mean deviation" term (exponential penalty term) that reflects the deviation from the center is comprehensively quantified, and its value directly corresponds to the stability of the process: the closer the value is to 1.0, the more stable it is, and the larger the value is, the worse the stability is; therefore, the output CI(t) is equivalent to generating a quality stability index.

[0184] The setting adopts the "normal distribution confidence interval and false alarm rate trade-off method", with a typical value range of [1.1, 1.5], and a preferred value of 1.2, corresponding to a safety margin of about 20%. Statistically, this usually covers 95% to 98% of the normal process fluctuation range. When CI(t)>1.2, it indicates that the current state has deviated significantly from the historical steady state, and there is a high probability that a substantial process variation has occurred (such as tool breakage or fixture loosening). Immediate intervention is necessary, and this is the best balance point between sensitivity and disturbance rejection.

[0185] The "balance method between the convergence and timeliness of the law of large numbers" is adopted for setting, with a typical value range of [20, 50], and a preferred value of 30. In statistical engineering, It is the dividing line between large and small samples. At this point, the distribution of the sample standard deviation is very close to the normal distribution, which can provide a stable benchmark reference. At the same time, 30 windows (if each window is 1 hour, that is, 30 hours of data) can cover multiple shifts to eliminate differences in personnel operation, and can also maintain sensitivity to recent process status. It is the best compromise between statistical reliability and dynamic adaptability.

[0186] The tolerance limit nonlinear mapping method is used for setting, with a typical value range of [1.5, 3.0] and a preferred value of 2.0. Utilizing the nonlinear amplification characteristic of the exponential function, when the mean offset reaches the tolerance half-width (critical non-compliance), This can amplify the exponential term to This ensures that any deviation close to the tolerance limit will trigger a mandatory alarm; while the amplification effect is mild for small deviations (<30% tolerance), avoiding excessive intervention.

[0187] S4. Construct a control objective function based on the production execution deviation set and quality stability index, and optimize the task allocation and processing parameters to generate a control vector;

[0188] S4.1: Normalize the set of production execution deviations and quality stability indicators, and construct the control objective function by weighted summation;

[0189] Specifically, based on the production execution deviation set The quality consistency index CI(t) is used to construct the comprehensive control objective function for closed-loop control. Since the cycle deviation, consistency index and load rate have different numerical scales, in order to avoid one of them dominating the objective function due to excessive scale, minimum-maximum normalization is performed. The control objective function is constructed with the goal of "reducing cycle deviation, bringing the load closer to the optimal point and reducing quality instability".

[0190] The control objective function is:

[0191]

[0192] In the formula, F(t) is the control objective function. It is the cycle time deviation of the s-th process. It is the load rate of the i-th device. This is the expected load level. , , These are weighting coefficients.

[0193] It should be noted that minimum-maximum normalization uses the maximum and minimum values ​​of each indicator obtained statistically within the historical stable operating cycle as boundaries to map the current real-time value to the [0,1] interval, eliminating the influence of dimensions, with the optimization goal of "reducing cycle deviation, bringing the load closer to the optimal point, and reducing quality instability".

[0194] The queuing time inflection point and buffer capacity trade-off method is adopted for setting, with a typical value range of [0.70, 0.80], and a preferred value of 0.75, which is at the optimal balance point between utilization and stability. It ensures a high equipment utilization rate of 75%, while reserving a flexible buffer space of 10% to 15% (a distance of 0.85 to 0.90 from the congestion line).

[0195] , , The settings are configured using the Analytic Hierarchy Process (AHP) combined with a dynamic production strategy mapping method, with a typical value range of [0.2, 0.6], and satisfying the following conditions: The preferred values ​​are 0.4, 0.3, and 0.3, respectively. The reason is that artillery production needs to take into account delivery schedule, equipment life balance, and quality stability, with schedule being slightly prioritized to ensure milestones are met. This enables the control system to adaptively adjust and optimize its focus based on order attributes, achieving a leap from "fixed rules" to "intelligent strategies".

[0196] S4.2: The control vector is generated using a hierarchical deterministic process of "task allocation first, then parameter fine-tuning".

[0197] Specifically, the first layer makes task allocation decisions: First, candidate equipment is screened based on hard lifespan constraints, and equipment with remaining lifespan less than the expected task duration is forcibly prohibited from undertaking new tasks; then, among the remaining equipment, task allocation variables are determined according to the rule of 'bottleneck priority load reduction and cycle time over-limit priority correction'.

[0198] The second layer performs fine-tuning of process parameters: branch logic is executed based on the status of the quality drift warning signal. If the warning signal is triggered, the quality forced constraint mode is activated to limit the feed rate to no more than the safety limit to force a speed reduction to ensure quality. If the warning signal is not triggered, the feed rate and spindle speed are fine-tuned within the allowable range of the conventional process to further optimize the objective function.

[0199] The final output is a control vector containing the task allocation scheme and parameter corrections. .

[0200] It should be noted that the meaning and constraints of each variable in the control vector are task-assigned variables. Feed rate correction factor Spindle speed correction factor ;

[0201] If task j is assigned to device i, then ,otherwise ;

[0202] Used for adjusting the CNC feed rate, based on the engineering technical limitations of the standard feed rate setting of the CNC system and the cutting stability lobe diagram. This range ensures that quality drift can be effectively suppressed by reducing speed, while avoiding drastic fluctuations in efficiency or plowing effect caused by excessive speed reduction.

[0203] Used for fine-tuning the spindle speed, based on engineering limitations of the spindle-tool system resonance characteristics and constant power / constant torque output curves. This range ensures that the adjustment is within the stable range allowed by conventional processes, avoiding large changes in rotational speed that could trigger system resonance.

[0204] A hard lifetime constraint prohibits allocation when the remaining lifetime of device i is insufficient to complete the task, i.e.:

[0205]

[0206] In the formula, It is the remaining lifespan. It is the estimated processing time for task j, which is obtained by summing the standard cycle times of each process in the process route and converting it into hours.

[0207] Quality constraint is a mechanism that forces a reduction in the feed rate to decrease processing disturbance when the quality consistency index exceeds a certain limit.

[0208]

[0209] In the formula, It is the mass drift threshold. It is the upper limit of the feed rate.

[0210] It should be noted that, Based on cutting stability theory and engineering emergency experience, the typical value is [0.75, 0.95], and the preferred value is 0.90. This value maps the standard emergency operation of experienced operators when they find an anomaly ("reduce the feed by 10% and see"). This implicit experience is made explicit as a safety threshold for automated control, ensuring that the system can automatically execute the most reliable and appropriate soft landing strategy when a quality alarm is triggered.

[0211] S5. Convert the control vector into equipment control commands and issue them for execution. Update the model parameters and control thresholds based on the execution feedback to form a continuous closed-loop control process.

[0212] S5.1: Convert the control vector into device control commands and issue them for execution;

[0213] Specifically, based on control vector Based on the task allocation variables, the target device for each task to be executed is determined, and a corresponding task scheduling instruction is generated. The task is written into the execution queue of the target device. Subsequently, the current feed rate of the CNC system is proportionally corrected according to the feed rate correction coefficient, and the spindle speed setpoint is adjusted accordingly according to the spindle speed correction coefficient. All instructions are sent to the corresponding device controller through a unified industrial communication protocol. After receiving the control instructions, the device parses and executes them, and returns an execution confirmation signal ACK. The execution status of all control instructions is recorded to form an execution confirmation status. When all key control instructions are successfully executed, the feedback analysis stage begins. If there are instructions that fail to execute, the relevant equipment or process is marked as an abnormal object and given priority in subsequent safety control.

[0214] S5.2: Based on the execution confirmation status and the updated field data, recalculate the cycle time deviation, equipment load rate, and quality consistency index to form a continuously iterative closed-loop control;

[0215] Specifically, the system collects and confirms the execution status and updated field data, recalculates cycle time deviation, equipment load rate, and quality consistency index to assess the actual effectiveness of control measures. Simultaneously, it verifies the calculated remaining lifespan by comparing it with the lifespan trend value obtained based on the latest health index trends, calculating the remaining lifespan error. When the error exceeds a set proportional threshold, an exponentially smoothed weighted method is used to update the degradation rate to ensure the lifespan model remains consistent with the actual degradation state, forming an updated model parameter set. When equipment load rate exceeds limits, quality consistency index significantly exceeds standards, or equipment remaining lifespan falls below the safety margin, a safety control mechanism is immediately triggered. After completing safety control, the current workshop status is updated, and the health index is recalculated based on the updated workshop status before entering the next control cycle.

[0216] The lifetime prediction error is:

[0217]

[0218] In the formula, It is a lifetime prediction error. It is the calculated remaining lifetime. It is a trend lifespan value estimated based on the latest rate of change in health indices;

[0219] The degradation rate parameter is corrected to:

[0220]

[0221] In the formula, This is the corrected degradation rate. It is the original degradation rate. The rate of degradation is calculated based on the latest health index differential. and It is a fixed weighting coefficient.

[0222] It should be noted that the proportion threshold is 10% of the calculated remaining lifetime. When the error is greater than 10% of the calculated remaining lifetime, it indicates that the original degradation model has deviated from reality.

[0223] Safety control mechanisms include reducing equipment operating speed, transferring tasks to equipment with lower load and sufficient lifespan, or suspending the current process to avoid quality control failure or sudden equipment failure.

[0224] Equipment load rate exceeding limits, quality consistency index seriously exceeding standards, or remaining equipment life less than the safety margin are as follows:

[0225]

[0226] In the formula, It is a safety margin.

[0227] The preferred value is 1.5 times the longest processing time of a single task or a fixed value of 4 hours (whichever is greater). This ensures that the equipment has sufficient margin for preventative maintenance or safe shutdown after completing the current task, avoiding workpiece scrapping due to sudden failures during processing.

[0228] and The "exponential smoothing filter and noise-response tradeoff method" is used for setting, with typical values ​​of [0.6, 0.9] and [0.1, 0.4], respectively. The preferred combination is 0.7 and 0.3, which is the best engineering compromise between "noise resistance" (preventing false alarms) and "sensitivity" (preventing missed alarms). It utilizes the stability of historical data to smooth out daily fluctuations, while retaining sufficient weight for new data to capture sudden accelerated wear trends, which meets the requirements of high reliability monitoring of artillery processing equipment.

[0229] This embodiment also provides a digital-based intelligent control system for artillery production workshops, including:

[0230] The data acquisition and alignment module is used to collect multi-source operating status and quality data, perform time alignment processing, and form a unified time-series workshop status data set;

[0231] The health life assessment module is used to calculate cycle time deviation and equipment load rate, and combined with remaining life constraints, to generate a set of production execution deviations containing multi-dimensional information;

[0232] The quality stability quantification module is used to calculate the quality consistency index based on key quality inspection parameters, comprehensively quantify quality fluctuations and mean deviations, and generate quality stability indicators.

[0233] The closed-loop optimization control module is used to construct the control objective function for optimization, generate control vectors and issue them for execution, and update parameters based on feedback to form a continuous closed loop.

[0234] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A digital-based intelligent management and control method for a gun production workshop, characterized in that: include, Multi-source operational status data and key quality inspection parameter data of processing equipment and workstations in the artillery production workshop are collected and time-aligned to form a unified time-series workshop status data set. Based on the workshop status data set, an equipment health index model based on digital twin residual and multi-dimensional feature adaptive fusion is constructed to obtain the equipment health vector. Based on recursive filtering and virtual extrapolation of future working conditions, the remaining life of the equipment is predicted to obtain the remaining life vector. The cycle time deviation is calculated based on the actual processing time and the preset standard cycle time. The equipment load rate is calculated by combining the task arrival rate and equipment capacity parameters. A set of production execution deviations is constructed by the remaining life vector of equipment, cycle time deviation, and equipment load rate. The quality consistency index is calculated based on key quality inspection parameter data as a quality stability indicator, thereby comprehensively quantifying quality fluctuations and mean deviations. Based on the set of production execution deviations and quality stability indicators, a control objective function is constructed, and the task allocation and processing parameters are optimized and solved to generate a control vector. The control vector is converted into equipment control commands and issued for execution. The model parameters and control thresholds are updated based on the execution feedback to form a continuous closed-loop control process. The steps for forming the device health vector and lifetime vector are as follows: Based on the workshop condition data set, vibration energy characteristics, impact characteristics, temperature rise trend characteristics, and absolute temperature characteristics were extracted and normalized. Based on normalized features, a digital twin reduced-order model is constructed to generate theoretical normal feature benchmark values ​​and calculate feature residuals. Based on feature residuals, a lightweight attention mechanism based on information entropy is used to adaptively update the weights and calculate the health index to obtain the device health vector. Based on the equipment health vector, the remaining lifetime vector is obtained by recursive filtering and virtual extrapolation of future operating conditions to predict the remaining lifetime. The production execution deviation set is generated by statistically calculating equipment load based on tasks and capabilities, marking status according to thresholds, and summarizing cycle time deviation, equipment load, and remaining usable life of the equipment into a production execution deviation set. The formation of continuous closed-loop control refers to converting control vectors into device control commands and issuing them for execution. Based on the execution confirmation status and the updated field data, the cycle time deviation, equipment load rate, and quality consistency index are recalculated to form a continuously iterative closed-loop control.

2. The digitization-based intelligent management and control method for a gun production workshop according to claim 1, characterized in that: The multi-source operating status data includes spindle speed, spindle current, vibration acceleration, bearing temperature, and actual processing time of a single piece. The key quality detection parameter data is the average value of the key quality parameters output by the detection station. The collected data is time-aligned to form a unified time-series set of workshop status data.

3. The digitization-based intelligent management and control method for a gun production workshop according to claim 2, characterized in that: The beat deviation is calculated based on the actual beat and the standard beat. Processes exceeding the threshold are marked as having a beat deviation exceeding the limit.

4. The digitization-based intelligent management and control method for a gun production workshop according to claim 3, characterized in that: The calculation of the quality consistency index based on key quality inspection parameter data as a quality stability indicator involves calculating the standard deviation and the quality consistency index, and triggering a quality drift warning signal when the threshold is exceeded.

5. The intelligent control method for a digitally based artillery production workshop as described in claim 4, characterized in that: The method of constructing a control objective function based on the production execution deviation set and quality stability index involves normalizing the production execution deviation set and quality stability index, and then constructing the control objective function by weighted summation.

6. The digitization-based intelligent management and control method for a gun production workshop according to claim 5, characterized in that: The optimization and solution of task allocation and processing parameters to generate control vectors is achieved by using a hierarchical deterministic process of first allocating tasks and then fine-tuning parameters.

7. An intelligent management and control system for a digitized artillery production workshop, based on any one of the intelligent management and control methods for a digitized artillery production workshop according to claims 1-6. include, The data acquisition and alignment module is used to collect multi-source operating status and quality data, perform time alignment processing, and form a unified time-series workshop status data set; The health life assessment module is used to calculate cycle time deviation and equipment load rate, and combined with remaining life constraints, to generate a set of production execution deviations containing multi-dimensional information; The quality stability quantification module is used to calculate the quality consistency index based on key quality inspection parameters, comprehensively quantify quality fluctuations and mean deviations, and generate quality stability indicators. The closed-loop optimization control module is used to construct the control objective function for optimization, generate control vectors and issue them for execution, and update parameters based on feedback to form a continuous closed loop.

Images