A dynamic spare parts inventory optimization method for a cigarette packing machine based on life difference perception

Abstract

The application discloses a kind of based on life difference perception's cigarette packet machine dynamic spare parts inventory optimization method, comprising: the data fusion and feature pre-processing of the multi-source heterogeneity of the key equipment of cigarette packet machine;Based on historical replacement cycle data, spare parts life is based on the explicit clustering of degradation mode;Construct life category driven differentiated demand forecasting submodel;Perform cost-sensitive safety inventory dynamic optimization mechanism;Perform online learning and strategy explainability enhancement module.The life difference perception-based cigarette packet machine dynamic spare parts inventory optimization method of the application, by explicitly identifying and classifying different life characteristics of spare parts, constructs differentiated prediction and inventory decision path, improves the pertinence of demand forecasting and safety inventory calculation, so as to more accurately balance the risk of stockout and holding cost, can solve the problem of inventory strategy misalignment caused by the life difference of the spare parts of the key equipment of the existing dynamic inventory method without explicitly modeling cigarette packet machine.

Description

Technical Field

[0001] This invention relates to the field of inventory optimization technology, and more specifically, to a method for optimizing the inventory of dynamic spare parts for tobacco packaging machines based on lifetime difference perception. Background Technology

[0002] Currently, cigarette packaging machines have numerous critical equipment components. Traditional safety stock management for critical equipment in cigarette packaging machines often uses static threshold settings, typically calculated based on historical average consumption and a fixed safety factor, without fully considering the lifespan differences between spare parts for critical equipment in different cigarette packaging machines. For example, critical components such as conveyor belts, sensors, and drive modules have significant differences in their design life, wear rate, and failure modes, resulting in large fluctuations in actual replacement cycles.

[0003] Static inventory strategies employ a uniform replenishment logic for all spare parts, which makes it difficult to adapt to the differentiated needs of high-life, low-frequency replacement parts and low-life, high-frequency consumable parts. This can easily lead to frequent shortages of short-life spare parts and long-term stockpiling of long-life spare parts, thereby increasing total holding costs and reducing service responsiveness.

[0004] While existing AI-based dynamic inventory methods incorporate multi-source data to predict demand, they generally treat spare parts lifespan as a latent variable and lack explicit modeling and classification decision-making mechanisms for lifespan characteristics. This leads to learning bias in the prediction model for weak correlations with lifespan, affecting the accuracy of inventory strategies.

[0005] Therefore, there is an urgent need for a dynamic spare parts inventory optimization method for tobacco packaging machines based on lifespan difference perception. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic spare parts inventory optimization method for cigarette packaging machines based on lifetime difference perception, in order to solve the problems in the prior art mentioned above. This method can improve the targeting of demand forecasting and safety stock calculation, thereby more accurately balancing stockout risk and holding costs.

[0007] This invention provides a method for optimizing dynamic spare parts inventory for tobacco packaging machines based on lifespan difference perception, comprising:

[0008] Data fusion and feature preprocessing are performed on the multi-source heterogeneity of key equipment in cigarette packaging machines;

[0009] Based on the data fusion and feature preprocessing results, and using historical replacement cycle data, explicit clustering of spare parts life based on degradation patterns is performed.

[0010] Based on the displayed clustering results of equipment lifespan, a lifespan-driven differentiated demand prediction sub-model is constructed;

[0011] Based on the aforementioned differentiated demand forecasting sub-model, a cost-sensitive safety stock dynamic optimization mechanism is implemented;

[0012] The differentiated demand prediction sub-model is optimized through online learning and strategy interpretability enhancement modules.

[0013] The dynamic spare parts inventory optimization method for tobacco packaging machines based on lifespan difference perception, as described above, preferably includes the following: data fusion and feature preprocessing of the multi-source heterogeneous components of key equipment in the tobacco packaging machine.

[0014] We collect full lifecycle operation data of key equipment in tobacco packaging machines. Through time alignment and event labeling strategies, we introduce a sliding window-based differential feature extraction method to calculate the rate of change of degradation rate of key equipment in the near replacement cycle. This rate is used as a latent variable input for early failure warning. Combined with the equipment OEE data of key equipment in tobacco packaging machines, we perform attribution filtering on abnormal downtime events.

[0015] The dynamic spare parts inventory optimization method for tobacco packaging machines based on lifespan difference perception, as described above, preferably includes the full lifecycle operation data of the key equipment of the tobacco packaging machine, including:

[0016] At least one of the following: spare parts model of key equipment of the cigarette packaging machine, installation time of key equipment, replacement record of key equipment, cumulative running time of key equipment, average daily start-up and shutdown frequency of key equipment, temperature and humidity conditions of the environment where key equipment is located, and vibration and current data of key equipment.

[0017] The above-described method for optimizing dynamic spare parts inventory for tobacco packaging machines based on lifespan difference perception, preferably, involves collecting full lifecycle operational data of key equipment in the tobacco packaging machine. Through time alignment and event labeling strategies, a sliding window-based differential feature extraction method is introduced to calculate the rate of change of degradation rate of key equipment near its replacement cycle. This rate serves as a latent variable input for early failure warning. Combined with the equipment OEE data of the key equipment in the tobacco packaging machine, attribution filtering of abnormal downtime events is performed, including:

[0018] The system integrates the SCADA system, MES work order records, equipment maintenance logs, and raw data streams from the edge acquisition gateway of the cigarette packaging machine. It collects the full lifecycle operation data of the key equipment of the cigarette packaging machine from the raw data streams. All full lifecycle operation data are uniformly sampled to a 5-minute granularity through a time alignment engine that applies satellite time synchronization and protocol synchronization methods, and the replacement cycle of each spare part is marked based on the start and stop events of the key equipment.

[0019] The degradation rate change rate is calculated using the sliding window differential method: with the vibration RMS sequence as input, a sliding window with a width of 24 hours is set in the last 7 days before replacement. The mean of the first difference of RMS within the window is calculated as the degradation rate, and then the second difference is calculated to obtain the degradation acceleration as the degradation rate change rate.

[0020] By combining OEE data analysis to attribute abnormal downtime: if the decrease in OEE during a certain replacement cycle is mainly caused by changeover debugging or material shortage rather than spare parts failure, the replacement event is marked as an atypical replacement and filtered out from the training set to ensure that each replacement cycle in the sample set truly reflects the physical wear and tear of spare parts.

[0021] The dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception, as described above, preferably includes the step of explicitly clustering spare parts lifespan based on degradation patterns according to historical replacement cycle data, based on data fusion and feature preprocessing results, comprising:

[0022] An improved spectral clustering algorithm is used to non-parametrically group the historical replacement cycles of various spare parts. The cycle mean, coefficient of variation and correlation between adjacent cycles are comprehensively considered as input feature vectors. A similarity metric function based on Weibull distribution prior is introduced. The optimal number of clusters is determined through adaptive contour analysis. Each spare part is assigned a clear physical semantic label. The physical semantic labels include: short-life high-frequency replacement, medium-life cycle replacement and long-life occasional replacement.

[0023] The above-described method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception, preferably, involves using an improved spectral clustering algorithm to non-parametrically group the historical replacement cycles of various spare parts. This grouping considers the cycle mean, coefficient of variation, and correlation between adjacent cycles as input feature vectors. A similarity metric function based on a Weibull distribution prior is introduced, and adaptive contour analysis is used to determine the optimal number of clusters. Each spare part category is then assigned a clear physical semantic label, including:

[0024] Based on the historical replacement cycle data of 47 types of spare parts, an input vector is constructed that includes three features: cycle length, coefficient of variation, and correlation coefficient between adjacent cycles of the historical replacement cycle data.

[0025] An improved spectral clustering algorithm is adopted, introducing a similarity metric function based on a three-parameter Weibull distribution, expressed by the following formula:

[0026] ;

[0027] in, Indicates the first Class of spare parts and the first Similarity measurement of spare parts Indicate duration The cumulative distribution function of the three-parameter Weibull distribution, This is a time variable, representing the service life of the spare part from installation. Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts. Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts;

[0028] The optimal number of clusters is determined by inputting a similarity metric into an adaptive contour analysis algorithm. Ultimately, all spare parts are divided into three categories: short-life high-frequency, medium-life cycle, and long-life occasional, so that each category of spare parts is given a clear physical semantic label.

[0029] The dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception, as described above, preferably includes the step of constructing a lifespan-category-driven differentiated demand prediction sub-model based on the displayed clustering results of equipment lifespan, comprising:

[0030] Design dedicated neural network structures and attention mechanisms for spare parts with different lifespan categories:

[0031] For short-life, high-frequency types, a time series regression model with gated cyclic units is constructed as the first differentiated demand prediction sub-model. The model focuses on modeling the nonlinear degradation path of cumulative runtime and dynamic load intensity, and introduces a periodic decay gating mechanism to simulate the wear accumulation effect.

[0032] For medium-life cycle components, a graph attention network is used, and a heterogeneous graph consisting of spare parts, their associated maintenance operations, and upstream and downstream components is used as the second differentiated demand prediction sub-model to learn the interaction between fault propagation paths and periodic maintenance.

[0033] For long-life, intermittent events, an anomaly triggering prediction model based on Transformer is deployed. The diagnostic logs of key equipment, transient current fluctuations, and potential fault modes labeled by experts are used as inputs. The model uses a self-attention mechanism to capture the precursor signs of rare strong signal events as the third differentiated demand prediction sub-model.

[0034] The dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception, as described above, preferably includes the step of implementing a cost-sensitive safety stock dynamic optimization mechanism based on the differentiated demand forecasting sub-model, which includes:

[0035] Each sub-model outputs the probability distribution of spare parts demand within the next T-period. Combining the Bayesian update mechanism with the uncertainty disturbances brought about by real-time operating condition changes, a single-period inventory decision model is constructed based on this distribution. The objective function comprehensively minimizes the expected shortage cost and holding cost per unit time. Among them, the shortage penalty cost is calculated by back-calculating the downtime loss of the key process it supports, while the holding cost considers capital occupation, warehouse depreciation, and the risk of expiration and scrapping. Differentiated service level constraints are set for three types of spare parts with different lifespans: short-life spare parts adopt a high service level to ensure continuous production, medium-life spare parts set a dynamic service level range that is adjusted according to the urgency of the work order, and long-life spare parts allow a lower basic service level but are equipped with an emergency procurement channel. The optimal safety stock threshold is generated in real time by solving a convex optimization problem using the Lagrange multiplier method.

[0036] The dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception, as described above, preferably includes optimizing the differentiated demand forecasting sub-model through an online learning and strategy interpretability enhancement module, comprising:

[0037] An online incremental learning framework is deployed in the online learning and policy interpretability enhancement module. When new replacement records or equipment upgrade data are injected into the system, model parameter fine-tuning and life category stability detection are automatically triggered. If a structural shift in the average life trend of a certain type of spare parts is detected, the re-clustering process is started and the corresponding sub-model structure is dynamically adjusted.

[0038] The dynamic spare parts inventory optimization method for cigarette packaging machines based on lifetime difference perception, as described above, preferably includes the following: Optimizing the differentiated demand forecasting sub-model through online learning and strategy interpretability enhancement modules further includes:

[0039] The system outputs recommended replenishment quantities and an interpretability report, which includes at least one of the following: ranking of dominant influencing factors, lifetime distribution evolution trend chart, and inventory strategy sensitivity analysis.

[0040] This invention provides a dynamic spare parts inventory optimization method for tobacco packaging machines based on lifespan difference perception. It collects spare parts types, historical replacement records, runtime, and operating condition data for key equipment in the tobacco packaging machine. Based on historical replacement cycle data, it performs lifespan clustering analysis on various spare parts, classifying them into short-life, high-frequency replacement, medium-life, and long-life, occasional replacement categories. For spare parts of different lifespan categories, it constructs customized feature engineering and demand prediction sub-models: the short-life model emphasizes operating load and cumulative runtime features; the medium-life model introduces the interaction term between fault frequency and maintenance cycle; and the long-life model adds weights for sudden fault signals and diagnostic alarms. Each sub-model outputs future demand prediction values. The system dynamically calculates the optimal safety stock threshold by matching the ratio of stockout penalty costs to holding costs for each category; it updates replenishment recommendations by category; by explicitly classifying spare parts by lifespan and implementing classification modeling, it improves the adaptability of the artificial intelligence model to different replacement patterns, making the dynamic inventory strategy more refined and interpretable; by explicitly identifying and classifying spare parts with different lifespan characteristics, it constructs differentiated forecasting and inventory decision-making paths, improving the targeting of demand forecasting and safety stock calculation, thereby more accurately balancing stockout risk and holding costs. This solves the problem of inaccurate inventory strategies caused by the lifespan differences of spare parts for key equipment in cigarette packaging machines that are not explicitly modeled in existing dynamic inventory methods. Attached Figure Description

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings, wherein:

[0042] Figure 1 This is a flowchart illustrating an embodiment of the dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception provided by the present invention. Detailed Implementation

[0043] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The descriptions of the exemplary embodiments are merely illustrative and are in no way intended to limit the present disclosure or its application or use. The present disclosure may be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided so that the present disclosure will be thorough and complete, and will fully express the scope of the disclosure to those skilled in the art. It should be noted that, unless specifically stated otherwise, the relative arrangement of components and steps, the composition of materials, numerical expressions, and values ​​set forth in these embodiments should be interpreted as exemplary only and not as limiting.

[0044] The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different parts. Terms such as “including” or “contains” mean that the element preceding the term encompasses the element listed after it, and do not exclude the possibility of encompassing other elements as well. Terms such as “above” and “below” are used only to indicate relative positional relationships; when the absolute position of the described object changes, this relative positional relationship may also change accordingly.

[0045] In this disclosure, when a specific component is described as being located between a first component and a second component, an intermediary component may or may not be present between the specific component and the first or second component. When a specific component is described as connecting to other components, the specific component may be directly connected to the other components without having an intermediary component, or it may not be directly connected to the other components but may have an intermediary component.

[0046] All terms used in this disclosure (including technical or scientific terms) have the same meaning as understood by one of ordinary skill in the art to which this disclosure pertains, unless otherwise specifically defined. It should also be understood that terms defined in a general dictionary, such as a dictionary, should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or highly formalized meaning, unless expressly defined herein.

[0047] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0048] Traditional safety stock management for spare parts in key tobacco packaging machines often employs static threshold settings, typically calculated based on historical average consumption and a fixed safety factor. This approach fails to adequately consider the varying lifespans of spare parts across different tobacco packaging machines. For instance, critical components such as conveyor belts, sensors, and drive modules exhibit significant differences in design lifespan, wear rates, and failure modes, leading to substantial fluctuations in actual replacement cycles. Static inventory strategies apply a uniform replenishment logic to all spare parts, making it difficult to adapt to the differentiated demands of high-lifespan, low-frequency replacement parts versus low-lifespan, high-frequency consumable parts. This can result in frequent shortages of short-lifespan spare parts and long-term stockpiling of long-lifespan spare parts, thereby increasing total holding costs and reducing service responsiveness. While existing AI-based dynamic inventory methods incorporate multi-source data to predict demand, they generally treat spare part lifespan as an implicit variable, lacking explicit modeling and classification decision mechanisms for lifespan characteristics. This leads to learning biases in the prediction models regarding weak correlations with lifespan, impacting the accuracy of inventory strategies.

[0049] This invention constructs differentiated forecasting and inventory decision-making paths by explicitly identifying and classifying spare parts with different lifespan characteristics, thereby improving the targeting of demand forecasting and safety stock calculation, and thus more accurately balancing stockout risk and holding costs.

[0050] like Figure 1 As shown in the figure, the dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception provided in this embodiment includes the following steps in actual implementation:

[0051] Step S1: Perform data fusion and feature preprocessing on the multi-source heterogeneity of key equipment in the cigarette packaging machine.

[0052] Specifically, the entire lifecycle operation data of key equipment in the cigarette packaging machine is collected. Through time alignment and event labeling strategies, a structured sample set with "spare parts-replacement cycle" as the basic unit is constructed. Each sample in the structured sample set contains the operating load intensity, environmental stress level, and abnormal alarm sequence of the components of the key equipment in the cigarette packaging machine within that cycle. A sliding window-based differential feature extraction method is introduced to calculate the rate of change of degradation rate of key equipment in the near replacement cycle, which serves as the latent variable input for early failure warning. Combined with the equipment OEE (Overall Equipment Efficiency) data of the key equipment in the cigarette packaging machine, abnormal downtime events are attributed and filtered to ensure that historical replacement records truly reflect spare parts life loss rather than external scheduling interference.

[0053] The full lifecycle operation data of the key equipment of the cigarette packaging machine includes at least one of the following: spare parts model of the key equipment, installation time of the key equipment, replacement record of the key equipment, cumulative running time of the key equipment, average daily start-up and shutdown frequency of the key equipment, temperature and humidity conditions of the environment where the key equipment is located, and vibration and current data of the key equipment.

[0054] Furthermore, the cigarette packaging machine involves a total of 47 categories of key spare parts, including: PLC main control module, servo driver power board, gearbox bearing assembly, industrial computer motherboard, network switch core module, pneumatic control valve solenoid coil, encoder interface board, temperature control relay, high voltage capacitor group, opto-isolation module, motor cooling fan controller, hydraulic unit pressure sensor, cutting blade, spindle end cover seal ring, spring pressure plate, guide wheel bushing, clamp rubber pad, clutch friction plate, cam follower bearing, solenoid valve sealing gasket, cylinder piston rod seal, etc. Synchronous belt tensioners, shear blades, photoelectric switch lenses, proximity switch sensors, encoder couplings, motor carbon brushes, drive chain link pins, filter elements, vacuum nozzles, lubrication nozzles, servo motors, photoelectric encoders, temperature probes, pressure transmitters, photoelectric sensors, limit switch contacts, relay contact groups, solenoid valve bodies, cylinder barrels, synchronous belts, chain guides, clamp cylinders, vacuum generator nozzles, thermocouple terminals, motor junction box seals, and control cabinet cooling fans can be categorized into four main types based on their function: transmission, control, execution, and sensing.

[0055] In one embodiment of the dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception of the present invention, step S1 may specifically include:

[0056] Step S11: Integrate the SCADA system, MES work order records, equipment maintenance logs, and raw data streams from the edge acquisition gateway of the integrated cigarette packaging machine, and collect the full lifecycle operation data of the key equipment of the cigarette packaging machine from the raw data streams.

[0057] All lifecycle operation data is uniformly sampled to a 5-minute granularity using a time alignment engine that applies satellite timing and protocol synchronization methods, and the replacement cycle of each spare part is marked based on the start and stop events of key equipment.

[0058] Taking the main drive camshaft seal ring of a key piece of equipment in a cigarette packaging machine as an example, the entire lifecycle operation data of the key equipment in the cigarette packaging machine from January 2021 was collected from the raw data stream. This included the installation batch number, replacement timestamp, cumulative running time (hours), average daily start-stop frequency, ambient temperature and humidity sensor readings, as well as the root mean square (RMS) value of vibration data per hour obtained through an accelerometer and load current data collected by a current transformer. All lifecycle operation data were uniformly sampled to a 5-minute granularity using a time alignment engine that applies satellite timing and protocol synchronization methods, and the replacement cycle of each spare part was marked based on the start-stop events of the key equipment. For example, a seal ring was installed on March 5, 2021, and replaced on April 12, 2021, due to a leakage alarm. Its service life was 38 days, and the cumulative running time was 912 hours. Within this period, the average daily start-stop frequency was extracted as 6.7 times / day, the average ambient temperature was 26.3℃, and the relative humidity was 62%.

[0059] Step S12: Calculate the rate of change of degradation rate using the sliding window differential method: Using the vibration RMS sequence as input, set a sliding window with a width of 24 hours in the last 7 days before replacement, calculate the mean of the first difference of RMS within the window as the degradation rate, and then calculate the second difference to obtain the degradation acceleration as the rate of change of degradation rate.

[0060] For example, on day 35, RMS increases from 0.82g to 0.98g, with a degradation rate of 0.016g / day; on day 36, it increases to 1.04g, with a degradation rate of 0.02g / day. Therefore, the degradation acceleration is 0.02 - 0.016 = 0.004g / day. 2 This metric is used as a latent variable input to subsequent models to capture early signs of performance degradation. Critical equipment and spare parts are of the same type.

[0061] Step S13: Analyze the causes of abnormal downtime by combining OEE data: If the decrease in OEE during a certain replacement cycle is mainly caused by changeover debugging or material shortage rather than spare parts failure, then the replacement event is marked as an atypical replacement and filtered out from the training set to ensure that each replacement cycle in the sample set truly reflects the physical wear and tear of spare parts.

[0062] Step S2: Based on the data fusion and feature preprocessing results, and based on historical replacement cycle data, perform explicit clustering of spare parts life based on degradation patterns.

[0063] Specifically, an improved spectral clustering algorithm is used to non-parametrically group the historical replacement cycles of various spare parts. The algorithm comprehensively considers the cycle mean, coefficient of variation, and correlation between adjacent cycles as input feature vectors. A similarity metric function based on the Weibull distribution prior is introduced to enhance sensitivity to the tail features of the lifespan distribution and avoid the bias of traditional Euclidean distance in long-tailed clustering. The optimal number of clusters is determined through adaptive contour analysis, and each spare part is assigned a clear physical semantic label. These physical semantic labels include: short-life, high-frequency replacement (e.g., sealing rings, cutting blades, replacement cycle <30 days), medium-life, periodic replacement (e.g., servo motors, photoelectric sensors, cycle 60–180 days), and long-life, occasional replacement (e.g., main control boards, gearboxes, cycle >2 years). The clustering results are used as explicit control variables in subsequent classification modeling and embedded into the prediction model architecture. In one embodiment of the present invention's dynamic spare parts inventory optimization method for cigarette packaging machines based on lifespan difference perception, step S2 may specifically include:

[0064] Step S21: Based on the historical replacement cycle data of 47 types of spare parts, construct an input vector containing three features: cycle length (days), coefficient of variation (CV), and correlation coefficient between adjacent cycles of the historical replacement cycle data.

[0065] Taking the cutting blade as an example, its historical replacement cycles were 28, 31, 25, 33, and 27 days, with an average cycle of 28.8 days, CV=0.103, and the correlation coefficient between adjacent cycles ρ=0.12, indicating that its replacement is relatively regular and is not significantly affected by previous use.

[0066] Step S22: An improved spectral clustering algorithm is adopted, introducing a similarity metric function based on a three-parameter Weibull distribution, expressed by the following formula:

[0067] ;

[0068] in, Indicates the first Class of spare parts and the first Similarity measurement of spare parts Indicate duration The cumulative distribution function of the three-parameter Weibull distribution, This is a time variable, representing the service life of the spare part from installation (in days or hours). Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts. Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts.

[0069] Step S23: Determine the optimal number of clusters by inputting the similarity metric into the adaptive contour analysis algorithm. Ultimately, all spare parts are divided into three categories: short-life high-frequency, medium-life cycle, and long-life occasional, so that each category of spare parts is given a clear physical semantic label.

[0070] Among them, the short-life, high-frequency type has a cycle of <30 days, including 19 types such as sealing rings, cutting blades, and spring pressure plates. The CV is generally lower than 0.15, and the degradation path is highly predictable.

[0071] The life cycle of medium-life products ranges from 60 to 180 days, including 16 types such as servo motor encoders, photoelectric sensors, and solenoid valves, with a CV between 0.18 and 0.35, which is greatly affected by fluctuations in operating conditions.

[0072] Long-life occasional type (>700 days): such as PLC main control module, servo drive, gearbox bearing, etc., a total of 12 types. Historical replacements were mostly caused by external impacts or software failures, not by natural aging.

[0073] Each type of spare part is assigned a clear physical semantic label and embedded as an explicit control variable in the predictive sub-model architecture to guide subsequent differentiated modeling strategies.

[0074] Step S3: Based on the displayed clustering results of equipment lifespan, construct a lifespan category-driven differentiated demand prediction sub-model.

[0075] Specifically, dedicated neural network structures and attention mechanisms are designed for spare parts with different lifespan categories:

[0076] For short-life, high-frequency classes, a time series regression model with gated cyclic units (GRU) is constructed as the first differentiated demand forecasting sub-model. The model focuses on modeling the nonlinear degradation path of cumulative runtime and dynamic load intensity, and introduces a periodic decay gating mechanism to simulate the wear accumulation effect.

[0077] For medium-life cycle components, Graph Attention Network (GAT) is used, which forms a heterogeneous graph of spare parts, their associated maintenance operations, and upstream and downstream components as the second differentiated demand prediction sub-model to learn the interaction between fault propagation paths and periodic maintenance.

[0078] For long-life, intermittent events, an anomaly triggering prediction model based on Transformer is deployed. The diagnostic logs of key equipment, transient current fluctuations, and potential fault modes labeled by experts are used as inputs. The model uses a self-attention mechanism to capture the precursor signs of rare strong signal events as the third differentiated demand prediction sub-model.

[0079] Among them, the three sub-models corresponding to the short-lifetime high-frequency class, the medium-lifetime periodic class, and the long-lifetime occasional class share the underlying feature encoder to achieve cross-category knowledge transfer, while retaining independent decision heads to ensure policy independence.

[0080] Step S4: Based on the differentiated demand forecasting sub-model, execute a cost-sensitive safety stock dynamic optimization mechanism.

[0081] Specifically, each sub-model outputs the probability distribution of spare parts demand within the next T-period. Combining this with a Bayesian update mechanism to incorporate the uncertainty caused by real-time operating condition changes, a single-period (S,s) inventory decision model is constructed based on this distribution. The objective function comprehensively minimizes the expected stockout cost and holding cost per unit time. The stockout penalty cost is calculated by back-calculating the downtime loss of the key processes it supports, while the holding cost considers capital occupation, warehousing depreciation, and the risk of expired scrapping. Differentiated service level constraints are set for three types of spare parts with different lifespans: short-life spare parts use a high service level (≥98%) to ensure continuous production; medium-life spare parts have a dynamic service level range (95%–97%) adjusted according to the urgency of work orders; and long-life spare parts allow a lower basic service level (90%) but are configured with an emergency procurement channel. The optimal safety stock threshold is generated in real-time by solving a convex optimization problem using the Lagrange multiplier method.

[0082] Step S5: Optimize the differentiated demand prediction sub-model through the online learning and strategy interpretability enhancement module.

[0083] Specifically, an online incremental learning framework is deployed in the online learning and strategy interpretability enhancement module. When new replacement records or equipment upgrade data are injected into the system, model parameter fine-tuning and life category stability detection are automatically triggered. If a structural shift in the average life trend of a certain type of spare parts is detected (e.g., Z-test p<0.01), the re-clustering process is started and the corresponding sub-model structure is dynamically adjusted.

[0084] Furthermore, step S5 also includes: the system outputting a recommended replenishment quantity and an interpretability report.

[0085] The interpretability report includes at least one of the following: a ranking of dominant influencing factors (e.g., the increase in forecasted demand for the current driving module is mainly due to a 23% increase in average daily runtime), a lifetime distribution evolution trend chart, and an inventory strategy sensitivity analysis. All decision-making logic supports traceability to original sensor data and replacement event logs, meeting the needs of industrial system auditing and continuous improvement.

[0086] The present invention provides a method for optimizing the dynamic spare parts inventory of tobacco packaging machines based on lifespan difference perception. This method collects data on spare parts types, historical replacement records, runtime, and operating conditions of key equipment in the tobacco packaging machine. Based on historical replacement cycle data, it performs lifespan clustering analysis on various spare parts, classifying them into short-life, high-frequency replacement, medium-life, and long-life, occasional replacement categories. For each lifespan category, customized feature engineering and demand prediction sub-models are constructed: the short-life model emphasizes operating load and cumulative runtime features; the medium-life model introduces the interaction term between fault frequency and maintenance cycle; and the long-life model adds weights for sudden fault signals and diagnostic alarms. Each sub-model outputs a predicted future demand value. By combining the ratio of stockout penalty cost to holding cost for each category, the optimal safety stock threshold is dynamically calculated. The system updates replenishment recommendations according to category. By explicitly classifying spare parts lifespan categories and implementing classification modeling, the adaptability of the artificial intelligence model to different replacement patterns is improved, making the dynamic inventory strategy more refined and interpretable. By explicitly identifying and classifying spare parts with different lifespan characteristics, differentiated prediction and inventory decision paths are constructed, improving the targeting of demand forecasting and safety stock calculation. This allows for a more accurate balance between stockout risk and holding costs, solving the problem of inaccurate inventory strategies caused by the lifespan differences of spare parts for key equipment in cigarette packaging machines that are not explicitly modeled in existing dynamic inventory methods.

[0087] The embodiments of this disclosure have now been described in detail. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

[0088] While specific embodiments of this disclosure have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments or equivalent substitutions can be made to some technical features without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifetime difference perception, characterized in that, include: Data fusion and feature preprocessing are performed on the multi-source heterogeneity of key equipment in cigarette packaging machines; Based on the data fusion and feature preprocessing results, and using historical replacement cycle data, explicit clustering of spare parts life based on degradation patterns is performed. Based on the displayed clustering results of equipment lifespan, a lifespan-driven differentiated demand prediction sub-model is constructed; Based on the aforementioned differentiated demand forecasting sub-model, a cost-sensitive safety stock dynamic optimization mechanism is implemented; The differentiated demand prediction sub-model is optimized through online learning and strategy interpretability enhancement modules.

2. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 1, characterized in that, The data fusion and feature preprocessing for the multi-source heterogeneous key equipment of the cigarette packaging machine includes: We collect full lifecycle operation data of key equipment in tobacco packaging machines. Through time alignment and event labeling strategies, we introduce a sliding window-based differential feature extraction method to calculate the rate of change of degradation rate of key equipment in the near replacement cycle. This rate is used as a latent variable input for early failure warning. Combined with the equipment OEE data of key equipment in tobacco packaging machines, we perform attribution filtering on abnormal downtime events.

3. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 2, characterized in that, The full lifecycle operational data of the key equipment in the cigarette packaging machine includes: At least one of the following: spare parts model of key equipment of the cigarette packaging machine, installation time of key equipment, replacement record of key equipment, cumulative running time of key equipment, average daily start-up and shutdown frequency of key equipment, temperature and humidity conditions of the environment where key equipment is located, and vibration and current data of key equipment.

4. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 2, characterized in that, The system collects full lifecycle operational data of key equipment in the tobacco packaging machine. Through time alignment and event labeling strategies, it introduces a sliding window-based differential feature extraction method to calculate the rate of change of degradation of key equipment near its replacement cycle. This rate serves as a latent variable input for early failure warning. Combined with the equipment OEE data of the key equipment in the tobacco packaging machine, it performs attribution filtering on abnormal downtime events, including: The system integrates the SCADA system, MES work order records, equipment maintenance logs, and raw data streams from the edge acquisition gateway of the cigarette packaging machine. It collects the full lifecycle operation data of the key equipment of the cigarette packaging machine from the raw data streams. All full lifecycle operation data are uniformly sampled to a 5-minute granularity through a time alignment engine that applies satellite time synchronization and protocol synchronization methods, and the replacement cycle of each spare part is marked based on the start and stop events of the key equipment. The degradation rate change rate is calculated using the sliding window differential method: with the vibration RMS sequence as input, a sliding window with a width of 24 hours is set in the last 7 days before replacement. The mean of the first difference of RMS within the window is calculated as the degradation rate, and then the second difference is calculated to obtain the degradation acceleration as the degradation rate change rate. By combining OEE data analysis to attribute abnormal downtime: if the decrease in OEE during a certain replacement cycle is mainly caused by changeover debugging or material shortage rather than spare parts failure, the replacement event is marked as an atypical replacement and filtered out from the training set to ensure that each replacement cycle in the sample set truly reflects the physical wear and tear of spare parts.

5. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 1, characterized in that, The step of explicitly clustering spare parts lifespan based on degradation patterns, based on historical replacement cycle data and the results of data fusion and feature preprocessing, includes: An improved spectral clustering algorithm is used to non-parametrically group the historical replacement cycles of various spare parts. The cycle mean, coefficient of variation and correlation between adjacent cycles are comprehensively considered as input feature vectors. A similarity metric function based on Weibull distribution prior is introduced. The optimal number of clusters is determined through adaptive contour analysis. Each spare part is assigned a clear physical semantic label. The physical semantic labels include: short-life high-frequency replacement, medium-life cycle replacement and long-life occasional replacement.

6. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 5, characterized in that, The improved spectral clustering algorithm is used to non-parametrically group the historical replacement cycles of various spare parts. It comprehensively considers the cycle mean, coefficient of variation, and correlation between adjacent cycles as input feature vectors. A similarity metric function based on a Weibull distribution prior is introduced, and the optimal number of clusters is determined through adaptive contour analysis. Each spare part category is then assigned a clear physical semantic label, including: Based on the historical replacement cycle data of 47 types of spare parts, an input vector is constructed that includes three features: cycle length, coefficient of variation, and correlation coefficient between adjacent cycles of the historical replacement cycle data. An improved spectral clustering algorithm is adopted, introducing a similarity metric function based on a three-parameter Weibull distribution, expressed by the following formula: ； in, Indicates the first Class of spare parts and the first Similarity measurement of spare parts Indicate duration The cumulative distribution function of the three-parameter Weibull distribution, This is a time variable, representing the service life of the spare part from installation. Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts. Indicate duration The The period length of the input vector of the spare part. Indicate duration The The coefficient of variation of the input vector of the spare part. Indicate duration The The correlation coefficient of the input vector of the spare parts; The optimal number of clusters is determined by inputting a similarity metric into an adaptive contour analysis algorithm. Ultimately, all spare parts are divided into three categories: short-life high-frequency, medium-life cycle, and long-life occasional, so that each category of spare parts is given a clear physical semantic label.

7. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 1, characterized in that, The step of constructing a lifespan-based differentiated demand prediction sub-model driven by equipment lifespan clustering results includes: Design dedicated neural network structures and attention mechanisms for spare parts with different lifespan categories: For short-life, high-frequency types, a time series regression model with gated cyclic units is constructed as the first differentiated demand prediction sub-model. The model focuses on modeling the nonlinear degradation path of cumulative runtime and dynamic load intensity, and introduces a periodic decay gating mechanism to simulate the wear accumulation effect. For medium-life cycle components, a graph attention network is used, and a heterogeneous graph consisting of spare parts, their associated maintenance operations, and upstream and downstream components is used as the second differentiated demand prediction sub-model to learn the interaction between fault propagation paths and periodic maintenance. For long-life, intermittent events, an anomaly triggering prediction model based on Transformer is deployed. The diagnostic logs of key equipment, transient current fluctuations, and potential fault modes labeled by experts are used as inputs. The model uses a self-attention mechanism to capture the precursor signs of rare strong signal events as the third differentiated demand prediction sub-model.

8. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 1, characterized in that, The step of implementing a cost-sensitive safety stock dynamic optimization mechanism based on the differentiated demand forecasting sub-model includes: Each sub-model outputs the probability distribution of spare parts demand within the next T-period. Combining the Bayesian update mechanism with the uncertainty disturbances brought about by real-time operating condition changes, a single-period inventory decision model is constructed based on this distribution. The objective function comprehensively minimizes the expected shortage cost and holding cost per unit time. Among them, the shortage penalty cost is calculated by back-calculating the downtime loss of the key process it supports, while the holding cost considers capital occupation, warehouse depreciation, and the risk of expiration and scrapping. Differentiated service level constraints are set for three types of spare parts with different lifespans: short-life spare parts adopt a high service level to ensure continuous production, medium-life spare parts set a dynamic service level range that is adjusted according to the urgency of the work order, and long-life spare parts allow a lower basic service level but are equipped with an emergency procurement channel. The optimal safety stock threshold is generated in real time by solving a convex optimization problem using the Lagrange multiplier method.

9. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 1, characterized in that, The optimization of the differentiated demand prediction sub-model through the online learning and strategy interpretability enhancement module includes: An online incremental learning framework is deployed in the online learning and policy interpretability enhancement module. When new replacement records or equipment upgrade data are injected into the system, model parameter fine-tuning and life category stability detection are automatically triggered. If a structural shift in the average life trend of a certain type of spare parts is detected, the re-clustering process is started and the corresponding sub-model structure is dynamically adjusted.

10. The method for optimizing dynamic spare parts inventory for cigarette packaging machines based on lifespan difference perception as described in claim 9, characterized in that, The optimization of the differentiated demand prediction sub-model through the online learning and strategy interpretability enhancement module also includes: The system outputs recommended replenishment quantities and an interpretability report, which includes at least one of the following: ranking of dominant influencing factors, lifetime distribution evolution trend chart, and inventory strategy sensitivity analysis.

Images