A rapid design method for an aero-engine part process

Abstract

The application relates to the technical field of electric digital data processing, and discloses a rapid design method for an aero-engine part process, which comprises the following steps: decomposing process knowledge into independent constraint atoms containing conditions and effects, and when reasoning, text matching is performed according to the effect state of one atom and the condition state of another atom, so that a one-way causal conduction path is constructed in real time, and then flow reasoning is completed through chain triggering. The application changes the organization and application mode of knowledge, flow modeling and reasoning are performed on the failure causal chain across the time and space scales, the system can penetrate the process barrier, delay failure caused by the historical process state is avoided, and mechanism basis is provided for tracing the failure analysis path.

Description

Technical Field

[0001] This invention relates to a rapid design method for the manufacturing process of aero-engine parts, belonging to the field of electronic digital data processing technology. Background Technology

[0002] Currently, a mainstream technical approach is to build a comprehensive process knowledge base or knowledge graph. Its core idea is to collect and aggregate a large number of discrete state points such as process parameters, equipment status, material properties and test results, and attempt to establish static mapping relationships between these states by using data association analysis or expert rules, so as to guide and optimize production.

[0003] However, when we place this state-aggregated data processing method in a typical manufacturing scenario spanning multiple processes, such as the complete process of a high-temperature alloy turbine disk from forging, heat treatment to final precision machining, its inherent, deeper limitations begin to emerge. The heat treatment process introduces complex, invisible residual stress fields inside the part. This stress field, as a key historical state, evolves slowly over time. When the part enters the high-speed precision machining process several days later, the existing process design system, when recommending cutting parameters, bases its decisions solely on the part's current geometric model, material grade, and other immediately visible state data. In terms of data processing paradigm, the system is inherently blind to the historical stress state that has long been lurking inside the workpiece and cannot be directly perceived by current sensors. As a result, a process decision that seems optimal in the current state may ultimately lead to delayed and catastrophic failures such as microcracks or deformation exceeding tolerances due to the superposition of invisible historical hidden dangers.

[0004] This phenomenon reveals an inherent mismatch between the fundamental paradigm of existing technologies in data processing and the physical essence of the causal temporality of manufacturing processes. It implicitly assumes that all key information influencing decisions is perceptible at the moment of decision-making, while neglecting the fact that many key factors have already solidified as historical states, and their corresponding effects erupt at distant points in time. Specifically, existing technologies suffer from the following shortcomings: 1. Limitations in the immediacy of information acquisition: Existing data processing methods excel at describing the current state but struggle to effectively incorporate historical latent information that has a decisive impact on future outcomes but cannot be directly measured; 2. Trans-temporal breaks in causal analysis: For processes separated by days or even weeks, existing technologies struggle to establish a logically coherent and mechanistically reliable long-term causal transmission chain between discrete state data points; 3. The static nature of knowledge representation: Complex process knowledge is solidified into a static network of relationships, making its query and reasoning patterns particularly cumbersome and inefficient when dealing with dynamically evolving causal processes.

[0005] Therefore, the technical problem to be solved by this invention is how to construct a streaming modeling and deduction method for failure causal chains that span time and space in the manufacturing process, so as to effectively avoid delayed failures caused by historical process conditions. Summary of the Invention

[0006] This invention provides a rapid design method for the manufacturing process of aero-engine parts. Its main purpose is to solve the problem that existing technologies, when performing digital process design, are unable to effectively trace and deduce the cause-and-effect chain of failures across processes and long cycles because their data processing methods are essentially focused on the current state and ignore historical timelines, thus making it impossible to avoid the risk of delayed failures.

[0007] To achieve the above objectives, this invention provides a rapid design method for the manufacturing process of aero-engine parts, the method comprising the following steps:

[0008] Step a, Define constraint atoms: Deconstruct the process specifications or known failure modes of aero-engine parts into multiple independently manageable digital constraint atoms, each containing a condition part and an effect part. Each constraint atom is defined as an IF-THEN logical judgment expression. The condition part describes one or more process states that need to be satisfied to trigger the constraint atom, and these process states are defined as one or more condition state labels that can be matched with text. The effect part describes the unique subsequent process impact generated after the constraint atom is triggered, and the subsequent process impact is defined as one or more effect state labels that can be matched with text.

[0009] Step b, Constructing a causal transmission path: Using the streaming simulation process as the method for executing process design, this method does not pre-establish a global static graph structure during process design. Instead, during the streaming simulation process, when the effect state label of one constraint atom matches the condition state label of another constraint atom, a unidirectional causal transmission path is established between the two constraint atoms in real time and implicitly.

[0010] Step c, perform streaming simulation: input the initial process conditions of the part to be designed to activate the initial constraint atoms that meet the corresponding condition state labels; then, broadcast the effect state label generated by each activated constraint atom as a new process state condition to trigger the matching constraint atoms on all subsequent causal transmission paths in a chain until a set of optimized process parameters, process parameter limits or failure risk warnings are output.

[0011] Preferably, in step a, each constraint atom is further associated with a weight value characterizing its constraint strength; and in the streaming derivation of step c, when the triggering conditions of two or more causal transmission paths are simultaneously satisfied and a conflict occurs, the method compares the weight values ​​of the corresponding constraint atoms on each conflicting path, executes the causal transmission path with the highest weight value, and suppresses the remaining causal transmission paths.

[0012] Preferably, the method further includes a step of elastically coordinating high-weight conflicts, which includes: when a conflict is detected in two or more causal transmission paths whose weight values ​​are all higher than a system-stored coordinating activation threshold, querying a pre-constructed process element correlation matrix; dynamically generating a temporary constraint atom containing boundary adjustment instructions based on the element with the highest correlation with the conflicting element in the process element correlation matrix; and injecting the temporary constraint atom into the subsequent streaming inference.

[0013] Preferably, the process element correlation matrix is ​​constructed by acquiring multidimensional process parameter data from a historical process database and calculating the statistical correlation between any two process parameters X and Y, wherein the statistical correlation is expressed as a correlation coefficient. The correlation coefficient is calculated according to the following relationship: in This represents the covariance of parameters X and Y. and Let X and Y represent the variances of parameters X and Y, respectively.

[0014] Preferably, the method further includes a step of associating constraint atoms with physical execution states and performing real-time tracing. This step includes: when defining the constraint atom, associating the constraint atom that produces a physical effect with a physical signal feature fingerprint extracted from sensor signals corresponding to its normal execution; acquiring physical sensor signals from the manufacturing process in real time and extracting their real-time signal features; when the deviation between the real-time signal features and the physical signal feature fingerprint of any constraint atom exceeds a predetermined range, locking the constraint atom as a potential anomaly source and adjusting its weight value in subsequent streaming inference.

[0015] Preferably, the physical signal feature fingerprint is the range of the main peak of the spectrum of the vibration signal in the predetermined frequency band during the processing, and the real-time signal feature is the main peak frequency of the spectrum extracted after the vibration signal is acquired in real time and subjected to a fast Fourier transform.

[0016] Preferably, the method further includes an adaptive compensation step for raw material batch differences, which includes: comparing the post-production quality inspection data of the first one or more parts in a specific production batch with the expected quality inspection results contained in the constraint atom to calculate one or more batch correction factors, the batch correction factors being used to characterize the material physical properties of the specific production batch; and applying the batch correction factors to dynamically correct the internal key parameters or their weight values ​​of one or more related constraint atoms when performing the flow cytometry simulation on all subsequent parts of the specific production batch.

[0017] Preferably, the method further includes a step of performing a systematic health diagnosis of the process system, which includes: continuously recording the conflicts that occur during the streaming simulation and converting the discrete conflict records into continuous time series data characterizing the frequency of conflict occurrence; performing frequency domain analysis on the time series data to identify the periodic patterns exhibited by the conflict occurrence; and generating diagnostic information indicating the existence of systemic risks in the process system based on the periodic patterns.

[0018] Preferably, the method further includes a step of actively detecting and adaptively warning of thermal stress abrupt changes during the quenching and cooling process. This step includes: acquiring temperature parameters of two key measuring points, the surface and the core, of the aero-engine part in real time during the quenching and cooling process; calculating a difference characterizing the instantaneous temperature gradient based on the two temperature parameters; and dynamically generating a temporary constraint atom when the difference exceeds a safety threshold related to the part material and stored in the system. The effect of the temporary constraint atom includes an instruction to reduce the flow rate of the cooling medium, and the temporary constraint atom is injected into the causal transmission path of the flow cytometry.

[0019] Preferably, step c further includes recording the sequence of all chain-triggered constraint atoms, and when outputting the failure risk warning, outputting the recorded sequence as a complete, traceable failure analysis path.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] The rapid design method for aero-engine parts disclosed in this invention has the following advantages:

[0022] 1. This invention deconstructs complex process knowledge into discrete constraint atoms containing conditions and effects. During the deduction process, a unidirectional causal transmission path is established only when the effect state of one constraint atom matches the condition state of another constraint atom. This changes the way process knowledge is organized and used. This modeling and deduction streaming architecture eliminates the need for the system to pre-build and maintain a global static graph structure. Instead, after inputting initial process conditions, the causal relationship can flow and evolve naturally along the path that meets the matching conditions until the final failure risk warning is output. This approach not only penetrates the barriers between processes and reveals the complete transmission path of delayed failures caused by historical process states, but also makes the construction and execution of the entire data processing flow logically simpler and more efficient.

[0023] 2. In the causal transmission path of the streaming simulation, this invention introduces a weight attribute to the constraint atom to characterize its constraint strength. When two or more high-weight causal transmission paths are satisfied simultaneously and conflict arises, it does not simply execute one and suppress the others. Instead, it treats this high-weight conflict itself as a specific event to activate a pre-constructed process element correlation matrix derived from the statistical correlation of historical process data. Based on the element with the highest correlation with the conflicting element in the matrix, the system dynamically generates a temporary constraint atom containing boundary adjustment instructions and injects it into the subsequent streaming simulation. This mechanism enables this method to find a flexible collaborative processing possibility from the physical correlation reflected by the data when facing rigid rule conflicts, thereby realizing the balancing and handling of complex process contradictions at a higher dimension.

[0024] 3. By associating constraint atoms that generate physical effects with a physical signal feature fingerprint corresponding to their normal execution, a direct mapping is established between abstract logical rules and the physical execution state in the manufacturing process. When the sensor signal features acquired in real time deviate from the fingerprint of any constraint atom, the system can directly lock the constraint atom as a potential source of anomaly and adjust its weight value in subsequent streaming simulations. This not only breaks down the information barrier between digital design and physical execution, enabling the correctness of process planning to be verified in real time by the actual manufacturing process, but also establishes an adaptive correction closed loop that feeds back from the physical world to the digital knowledge system, allowing the process knowledge itself to iterate and evolve based on practical results. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the logical architecture and execution process of the present invention.

[0026] Figure 2 This is a schematic diagram illustrating the active detection of sudden changes in thermal stress during the quenching and cooling process of this invention.

[0027] Figure 3 This is a flowchart illustrating the closed-loop application process of the entire manufacturing process of this invention.

[0028] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0030] The core technology of the rapid design method for aero-engine parts disclosed in this invention lies in an electronic digital data processing procedure that deconstructs, transmits, and extrapolates process knowledge. This procedure aims to construct a dynamic processing architecture for parts such as high-temperature alloy turbine disks, which are prone to delayed failures in multi-process, long-cycle manufacturing processes, enabling streaming modeling and extrapolation of failure causal chains across time and space scales. The method mainly includes three core steps: defining constraint atoms, constructing causal transmission paths, and performing streaming extrapolation. Through a logical flow of modeling and extrapolation, the system can penetrate process barriers, proactively reveal and avoid delayed failures caused by historical process conditions, and also provide a clear mechanistic basis for tracing the complete failure analysis path.

[0031] In a specific application scenario, such as digitally designing the entire process of a high-temperature alloy turbine disk from forging and heat treatment to final precision machining, a common technical challenge is that early processes such as heat treatment introduce complex and invisible residual stress fields inside the part. Subsequent process decision-making systems, due to their data processing paradigm being blinded to this already solidified historical state, often make process decisions that seem optimal at the moment, but which ultimately overlap with historical hidden dangers, leading to microcracks or excessive deformation. To address this challenge, the method disclosed in this invention first performs step a, namely defining constraint atoms. This step deconstructs the process specifications or known failure modes of aero-engine parts into multiple independently manageable digital constraint atoms, each containing conditional and effect parts. Each constraint atom is a logical judgment expression, and its conditional part describes one or more process states that need to be satisfied to trigger the constraint atom. The effect section describes the unique subsequent process impact generated after the atom is triggered. Both the condition and effect sections contain process status labels that can be matched with text. For example, a heat treatment constraint can be defined as atom A: IF(Material=GH4169ANDHeatTreatment_Temp>1100°C)THENSET(InternalStress_Risk=HIGH), while a cutting constraint can be defined as atom B: IF(InternalStress_Risk=HIGHANDCutting_Depth>0.5mm)THENTRIGGER(Crack_Failure_Alert). By deconstructing complex process knowledge into discrete, logically self-consistent constraint atoms, this method lays a flexible and independently manageable knowledge representation foundation for the subsequent construction of dynamic causal chains.

[0032] Given that static knowledge graphs struggle to establish logically coherent long-term causal transmission chains across processes spanning several days, the system employs step b to construct causal transmission paths. This method does not pre-build a global static graph structure. Instead, during the streaming deduction process, when the effect state label of a constraint atom, such as InternalStress_Risk=HIGH, achieves a precise text match with the condition state label of another constraint atom, a unidirectional causal transmission path is established instantaneously and implicitly between these two constraint atoms. This dynamic path construction method, which relies on effect-condition text matching, allows knowledge association and transmission to proceed without relying on any predefined graph structure, thereby reducing the construction and maintenance costs of the knowledge base.

[0033] Based on this, the system executes step c, namely, the streaming deduction. When the initial process conditions of the part to be designed are input, such as the material being GH4169 and the heat treatment temperature being 1150°C, this input will be used as initial broadcast information to activate all initial constraint atoms that meet the corresponding condition status labels. Subsequently, the effect status label generated by each activated constraint atom, such as InternalStress_Risk=HIGH, will be broadcast again as a new process state condition to trigger all matching constraint atoms on the subsequent causal transmission path in a chain, until the system traverses all triggerable paths and outputs the final process parameter limit or failure risk warning. In addition, during this deduction process, the sequence of all chain-triggered constraint atoms will be recorded. When outputting the failure risk warning, the recorded sequence will be output as a traceable complete failure analysis path, thus providing process designers or failure analysts with a logically clear and mechanistically reliable decision-making basis.

[0034] In complex process scenarios, multiple process constraint paths may be satisfied simultaneously, leading to decision conflicts. To make deterministic decisions on such conflicts, in step a, each constraint atom is associated with a weight value representing its constraint strength. Correspondingly, in the streaming derivation in step c, when the triggering conditions of two or more causal transmission paths are satisfied simultaneously, this method compares the weight values ​​of the corresponding constraint atoms on each conflicting path, executes the causal transmission path with the highest weight value, and suppresses the remaining causal transmission paths, thereby providing a simple and efficient priority judgment mechanism for the system's decision logic.

[0035] However, for some high-value parts operating under extreme conditions, simple either-or rigid decisions may miss the overall collaborative processing solution. Therefore, this method also includes a step for flexible collaborative processing of high-weight conflicts. This step first sets a collaborative activation threshold stored in the system. When a conflict is detected in two or more causal transmission paths where both weight values ​​are higher than this collaborative activation threshold, the system does not immediately execute a rigid decision. Instead, it queries a pre-constructed process element correlation matrix. This process element correlation matrix is ​​obtained by acquiring multi-dimensional process parameter data from a historical process database and calculating the correlation between any two process parameters. and It is constructed based on the statistical correlation between them, wherein the statistical correlation is expressed as the correlation coefficient. The correlation coefficient is calculated according to the following relationship: in Indicates parameters and The covariance, and and Representing parameters respectively and Based on the variance of the process element correlation matrix, the system dynamically generates a temporary constraint atom containing boundary adjustment instructions and injects it into the subsequent streaming deduction. This allows the system to find a possible collaborative solution from the physical correlation reflected in the data when facing rigid rule conflicts.

[0036] To break down the information barrier between digital design and physical execution, this method also includes a step of associating constraint atoms with physical execution states and performing real-time traceability. When defining constraint atoms, this step associates the constraint atom that produces a physical effect with a physical signal feature fingerprint extracted from sensor signals corresponding to its normal execution. For example, this physical signal feature fingerprint can be determined as the spectral peak range of the vibration signal within a predetermined frequency band during processing. This range can be calibrated by collecting vibration signals from test-cut parts under standard process conditions, extracting the statistical distribution range of the spectral peak after Fast Fourier Transform (FFT), and so on. In actual production, the system acquires physical sensor signals from the manufacturing process in real time and extracts their real-time signal features, such as the spectral peak frequency extracted from the real-time vibration signal after FFT. When the real-time signal features deviate from the physical signal feature fingerprint of any constraint atom beyond a predetermined range, the constraint atom is identified as a potential anomaly source, and its weight value in subsequent streaming simulations can be adjusted accordingly, thus establishing an adaptive correction closed loop that feeds back from the physical world to the digital knowledge system.

[0037] Considering that batch variations of raw materials are a significant factor leading to process instability, this method also includes an adaptive compensation step for batch variations of raw materials. This step compares the post-production quality inspection data of the first one or more parts in a specific production batch with the expected quality inspection results contained in the constraint atoms to calculate one or more batch correction factors characterizing the physical properties of the materials in that batch. When performing flow cytometry simulations on all subsequent parts in that batch, the system automatically applies these batch correction factors to dynamically correct the internal key parameters or their weight values ​​of one or more related constraint atoms, thereby enabling process decisions to intrinsically and automatically compensate for the unique physical properties of the materials in the current batch.

[0038] To diagnose systemic risks in a process system, this method also includes a step of performing a systemic health diagnosis of the process system. This step continuously records conflict events that occur during the streaming simulation and converts the discrete conflict event records into continuous time series data characterizing the frequency of conflict occurrence. Subsequently, frequency domain analysis is performed on the time series data to identify the periodic patterns exhibited by the conflict events. Based on these periodic patterns, the system can generate diagnostic information indicating the existence of systemic risks in the process system, thereby elevating decision support from solving single problems to identifying and eradicating systemic problems.

[0039] To address the challenge of predefining thermal stress abrupt changes under extreme conditions such as quenching and cooling, this method also includes a step for active detection and adaptive early warning of thermal stress abrupt changes. This step acquires the temperature parameters of two key measuring points, the surface and core, of the aero-engine parts in real time during the quenching and cooling process, and calculates a difference representing the instantaneous temperature gradient based on these two temperature parameters. This difference is compared with a safety threshold stored in the system based on the part material. When the difference exceeds the safety threshold, the system dynamically generates a temporary constraint atom, the effect of which includes an instruction to reduce the flow rate of the cooling medium. This temporary constraint atom is then injected into the causal transmission path of the flow cytometry. Thus, the system can make a real-time, closed-loop adaptive response to unpredictable physical abrupt changes that actually occur during the manufacturing process.

[0040] Example 1: In the actual production process of a high-temperature alloy turbine disk for an aero-engine, when a specific batch of high-temperature alloy material is used, and after it has completed the forging and heat treatment processes, it enters the final high-speed precision machining process. The machining of the first part of the batch strictly follows the existing process document parameters. However, the part was still found to have delayed microcracks caused by the superposition of residual stress and processing stress in the subsequent non-destructive testing. This result led to the scrapping of the part and the interruption of the production plan.

[0041] To address this type of failure caused by the superposition of historical latent factors and current processes, the technical solution of this invention is applied to the process design of subsequent parts in the same batch. First, the system inputs the complete process data of the first scrapped part, the subsequent metallographic analysis data, and the non-destructive testing results as a set of data containing expected and actual results. By comparing the systematic deviations between the expected hardness and grain size and the actual test values, the system calculates a batch correction factor for the physical properties of the material in this batch according to the procedures described in the specific implementation. This factor is then used to dynamically correct the weight values ​​of the constrained atoms related to the material's heat treatment history. In this way, the logical starting point for all subsequent deductions has been embedded with compensation for the physical properties of the material in this specific batch.

[0042] When planning the process parameters for the second part in this batch, flow cytometry was initiated. The InternalStress_Risk effect, generated after the activation of the heat treatment history-related constraint atoms adjusted by the batch correction factor, had a weight value significantly higher than the standard value. When this effect state conflicted with a high-weight constraint atom in the high-speed cutting process designed to ensure surface integrity in the causal transmission path, the system detected that the weight values ​​of both causal transmission paths were higher than the preset co-activation threshold. At this point, the system did not perform a rigid binary choice but instead triggered a query of the process element correlation matrix. The query results showed that, in the historical data, there was a value between the two process parameters, cutting depth and cooling medium flow rate, with a value... Based on the statistical correlation, the system dynamically generates a temporary constraint atom containing a command to moderately increase the cooling medium flow rate, and injects it into the subsequent flow cytometry simulation.

[0043] After injecting temporary constraint atoms, the system performed another simulation. The new causal transmission path showed that, with optimized cooling conditions, the original cutting depth limit could be appropriately relaxed within the boundary without inducing the risk of microcracks. The system finally output a set of co-optimized process parameters and guided the machining of the second part. After machining, the part passed all non-destructive testing and no signs of microcracks were found. Its dimensional accuracy and surface integrity indicators also met the design requirements. Based on this, all subsequent parts using this batch of materials were produced using this verified causal transmission path and parameter set, showing stable and consistent machining results.

[0044] This method logically links discrete batch posterior data, statistical correlations of historical data, and real-time extrapolation for specific parts, transforming the originally static and isolated process knowledge into a dynamic decision flow that can adaptively evolve according to specific working conditions. Instead of attempting to build an all-encompassing static knowledge base, it generates a logically consistent and verifiable causal transmission path for each part to be processed in real time, thereby transforming the digital control of complex manufacturing processes from state-based static mapping to causal time-series extrapolation.

[0045] Example 2: To quantitatively verify the effectiveness of the present invention in associating abstract logical rules with the physical execution state in the manufacturing process and performing real-time traceability, this comparative experiment was designed. The purpose of the experiment is to objectively prove that by associating a physical signal feature fingerprint with a constraint atom, the system can attribute abnormal deviations of physical signals to their corresponding logical sources, i.e., specific constraint atoms, in a simulated, progressive physical failure scenario, and adjust subsequent streaming inference accordingly.

[0046] The experiment was conducted on a CNC machining center equipped with a high-frequency vibration sensor, which was fixed to the tool clamping system to collect real-time vibration signals during the machining process. The workpiece used was a high-temperature alloy standard test block. The experiment was divided into a control group and an embodiment group of the present invention. Both groups performed the same continuous cutting task defined by a series of constraint atoms and used the same tool that would produce progressive wear to simulate the real failure process. The control group used a traditional equipment health monitoring system, which only monitored the overall amplitude of the vibration signal and set a static alarm threshold. The embodiment group of the present invention used the technical solution of the present invention. The sampling frequency of the vibration signal was set to 10 kHz. The technical factors balancing this setting were to satisfy the Nyquist sampling theorem to capture the characteristic frequency of the tool under different wear states, which may be as high as 4 kHz, without distortion, and to take into account the real-time computing load of the data processing unit. The sampling frequency of 10 kHz is 2.5 times the estimated highest characteristic frequency, providing sufficient fidelity for signal analysis.

[0047] In the experimental preparation phase of the present invention embodiment group, firstly, for a core cutting constraint atom, the effect part of which includes a specific set of cutting parameters, 10 standard cuts are performed using a brand new, healthy tool, and the vibration signal during each cut is subjected to a fast Fourier transform to extract its spectral main peak frequency. Based on the mean and standard deviation of these 10 sample data points, a confidence interval is calculated, which is defined as the physical signal feature fingerprint of this constraint atom. After the experiment begins, as the tool wears progressively, the monitoring system of the control group triggers a shutdown alarm when the total vibration amplitude exceeds a preset threshold when processing the 8th test block, but does not provide any directional information about the root cause of the abnormality. In the present invention embodiment group, the system continuously compares the spectral main peak frequency of the real-time vibration signal with the calibrated physical signal feature fingerprint.

[0048] Table 1 shows some status monitoring and response data during the processing. In the early stage of the processing, the main peak frequency of the real-time spectrum fluctuated slightly within the fingerprint range, and the system judged the status to be normal. When the processing reached the fifth test block, the main peak frequency of the real-time spectrum began to deviate continuously from the upper limit of the fingerprint range. When the deviation exceeded the predetermined range, the system locked the cutting constraint atom as a potential anomaly source and adjusted its weight value in the subsequent streaming simulation in real time and slightly. This adjustment caused the system to automatically give a more conservative combination of cutting parameters when planning the process for the sixth test block. This process shows that the method of the present invention does not passively wait for the physical quantity to exceed the hard limit, but rather detects the deterioration trend of the process execution status in advance by tracking the drift of the feature fingerprint.

[0049] Table 1: This table shows some status monitoring and response data of the embodiment group of the present invention during continuous processing.

[0050]

[0051] Experimental results show that the method disclosed in this invention constructs a direct feedback link from the physical execution site to the digital knowledge system by establishing a feature fingerprint for the constraint atoms at the logical level that can be verified by physical sensor signals. This mechanism enables the system not only to detect the occurrence of physical anomalies, but also to probabilistically attribute them to specific process design links, i.e., a certain constraint atom. This transforms the response to anomalies from a post-event, non-directional alarm to a real-time, attribution-capable adaptive adjustment, providing a feasible technical path for realizing the closed-loop iteration and evolution of process knowledge in practice.

[0052] Example 3: This example combines Figures 1 to 3 This document describes a rapid design method for the manufacturing process of an aero-engine component, such as... Figure 1 As shown, the process begins with initial process condition input, which includes the initial state of the part to be designed, such as material and geometry. This input is fed into a constraint atom library that deconstructs process specifications and failure modes into independent condition-effect atoms. This atom library can adaptively adjust its weights based on feedback from the physical execution state correlation and real-time traceability. The core of the process is a streaming inference engine, which dynamically constructs causal paths by matching the effect of one atom with the condition of another atom using text. If no decision conflict occurs during the inference, the final result is directly output, which includes process parameter limitations / failure risk warnings and a complete, traceable failure analysis path. If a decision conflict occurs... The system then determines whether the conflict is a high-weight conflict. If not, it executes the path with the highest weight and outputs the result. If it is, it activates a dynamic collaboration and feedback closed-loop mechanism. This closed-loop mechanism includes two core components. The first is the elastic collaborative processing of high-weight conflicts. It dynamically generates temporary constraint atoms containing boundary adjustment instructions by querying a pre-built process element correlation matrix and injects them into the backflow inference engine to handle the conflict. The second is the physical execution state correlation and real-time traceability. It receives external physical sensor signals, compares real-time signal characteristics with preset physical signal feature fingerprints, and locks the abnormal source when the signal deviates and corrects its atomic weights, thereby feeding back the execution state of the physical world to the digital knowledge system.

[0053] like Figure 2As shown in the figure, the horizontal axis represents the cooling time in seconds, and the vertical axis represents the surface-core temperature difference in °C. A dashed line in the figure represents a constant safety threshold, while a solid line represents the temperature gradient in °C monitored in real time during the quenching process. At about 25 seconds, the measured temperature gradient curve shows a sharp abrupt change, and its peak briefly exceeds the preset safety threshold. This exceedance event is a physical abrupt change that can be actively detected by the system. This event will trigger the system to dynamically generate a temporary constraint atom containing specific instructions and inject it into the causal transmission path of the streaming inference to achieve adaptive early warning and response to thermal stress abrupt changes.

[0054] like Figure 3 As shown, the process begins with inputting initial process conditions, driving the part to be processed into the process planning process, and initiating process simulation. During the simulation, latent high internal stress risks introduced by historical processes such as heat treatment may trigger state-process decision conflicts after matching with subsequent cutting atomic conditions. At this time, the system enters the state of collaborative optimization, adjusts the boundary by generating temporary constraint atoms, and sends the optimization results back to the main simulation process. After the simulation is completed, if a high risk is identified, a failure risk warning is output; if there is no risk, the part enters actual processing. During processing, physical state verification is performed by comparing real-time sensor signals with fingerprints. If the real-time signal is out of range, it is determined that the state physical signal deviates. This event will trigger the locking of the abnormal source / adjustment of the atomic weights on the one hand, and will also feed this information back to the collaborative optimization for dynamic adjustment on the other hand. If the physical verification is passed, the process will eventually move towards process solidification / qualification, thus forming a complete adaptive manufacturing closed loop that integrates latent state warning, decision conflict coordination, physical state verification, and dynamic feedback.

[0055] Example 4: In a specific deployment scenario, in order to apply the general method described in this invention to a specific aero-engine parts production line, a series of offline calibration and parameter determinism procedures need to be performed to ensure that all decisions and inferences of the system are based on quantifiable engineering basis that matches the specific production line and parts family.

[0056] For the elastic collaborative processing mechanism, the determination of the collaborative activation threshold, which is the basis for triggering the mechanism, follows a calibration procedure based on the statistical distribution of historical conflict events. This procedure first extracts all decision conflict events recorded by the system during historical operation or simulation, and for each conflict event, records the weight values ​​of the corresponding constraint atoms on the two causal transmission paths that conflict with each other when they occur. Subsequently, all weight value data pairs are statistically analyzed, and their two-dimensional joint probability distribution is plotted. The collaborative activation threshold is set on the boundary of a high-weight and low-probability region in this distribution. A feasible setting method is to select the lowest weight value corresponding to the event whose weight value is in the top 5% of all conflict events as the threshold value. This procedure anchors the setting of the collaborative activation threshold to the distribution of high-weight and low-probability conflict events in historical data, so that the subsequent elastic collaborative processing mechanism can be reserved for handling major decision conflict scenarios.

[0057] For the active detection and adaptive early warning mechanism of thermal stress abrupt change during quenching and cooling, the core judgment criterion, namely the safety threshold related to the part material, is determined through a process combining material physical properties and finite element analysis. This process first obtains the yield strength data of the high-temperature alloy used in the part at different temperatures from a standard material property database, forming a curve showing the yield strength versus temperature. Second, a one-time, simplified thermo-structure coupled finite element analysis is performed on the geometric model of the part to establish the instantaneous temperature gradient between the part's surface and core. Maximum thermal stress inside the part The approximate functional relationship between them; ultimately, at any real-time measured core temperature point, the safety threshold used by the system. The value of this value is determined such that the maximum thermal stress calculated from this approximate functional relationship is obtained. The yield strength of the material at the core temperature shall not exceed a preset safety factor, which may be 70%. Through this procedure, a static threshold concept is transformed into a calculable function that is dynamically related to the real-time state of the material.

[0058] Similarly, in the process of systematically diagnosing the health of the process system, the algorithmic path for converting discrete conflict event records into continuous time series data characterizing the frequency of conflict occurrence is as follows: The system first defines a time window. and a step increment At each time step point The system is in Count all conflict events that occur within the time window to obtain a count value. This series of SN(t)S forms a continuous time series data that can be used for frequency domain analysis on the time axis; this algorithm path transforms unstructured log records into structured digital signals that can be mathematically analyzed.

[0059] Example 5: Before deploying the framework described in this invention in a specific industrial production environment, an offline data filling and model building procedure needs to be executed to generate the process element correlation matrix described in the previous embodiments. This procedure first extracts process parameter records and corresponding quality inspection results from thousands of complete work orders covering various part types and processing conditions from a historical production database. The original dataset is then cleaned and normalized to eliminate outliers and dimensional influences. Subsequently, the system iterates through all predefined key process parameter pairs, which include depth of cut, spindle speed, feed rate, and cooling medium flow rate, and calculates the relationships based on the statistical correlations disclosed in the specific embodiments. The correlation coefficient between any two process parameters is calculated. Finally, all these correlation coefficients are organized into an N×N symmetric matrix. This matrix is ​​fixed in the system as a static, version-controlled digital object, providing a quantifiable decision-making basis that reflects the inherent physical laws of the production line for high-weight conflict events that may occur in subsequent streaming simulations and require flexible collaborative processing.

[0060] For batch correction factors used to handle batch variations in raw materials, the system employs a continuously running, dynamic knowledge base construction and retrieval procedure. This procedure establishes a structured database with material suppliers, material grades, and production batch numbers as joint primary keys. When a new batch of materials is put into production, the system compares the post-production quality inspection data of the first or first few parts of that batch with the baseline expected value, calculates one or more batch correction factors bound to the characteristics of that batch, and stores these factors along with the batch information in the database. When processing any subsequent part of that batch, the streaming simulation system first queries the database based on the batch number of that part to obtain the corresponding correction factor. Before the simulation begins, it applies this factor to dynamically correct the internal parameters or weight values ​​of one or more related constraint atoms. This procedure transforms the response to raw material uncertainties from a one-off, passive adjustment into a continuously accumulating, proactively adaptable knowledge management process.

[0061] Example 6: To ensure the effectiveness and completeness of the constraint atoms and their weight values, which serve as knowledge carriers in the method described in this invention, during long-term operation, the system is configured to execute a periodic, offline knowledge base health diagnosis and optimization procedure. This procedure first retrieves the streaming deduction process records of all completed parts within a preset evaluation period from the production database. These records contain the quality inspection data of each activated causal transmission path and its final association. The system performs statistical analysis on these data, calculates the activation frequency of each constraint atom, and identifies those constraint atoms that are never or rarely activated in actual operation. These are marked as potential redundant rules for review by engineering technicians.

[0062] Furthermore, the procedure also performs correlation analysis between all causal transmission paths that trigger failure risk warnings and actual part scrapping events. If a warning path is frequently triggered, but its corresponding actual failure rate is lower than a preset confidence threshold, the system will reduce the weight of key constraint atoms on that path according to a predetermined function based on the path's historical false alarm rate. Conversely, if no high-weight causal transmission path is activated in the system before a certain type of part failure event occurs, the system will collect the data of these predicted failure cases and perform cluster analysis on all process parameters of these cases using an unsupervised learning algorithm to discover previously undefined, hidden failure mode combinations. If the algorithm identifies a clear parameter combination that is strongly correlated with a specific failure mode, the system will submit this correlation as a hypothetical, new constraint atom to engineering technicians for mechanism verification and confirmation. After confirmation, the new constraint atom will be formally added to the knowledge base.

[0063] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0064] Finally, 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.

Claims

1. A rapid design method for the manufacturing process of aero-engine parts, characterized in that, The method includes the following steps: Step a, Define constraint atoms: Deconstruct the process specifications or known failure modes of aero-engine parts into multiple independently manageable digital constraint atoms, each containing a condition part and an effect part. Each constraint atom is defined as an IF-THEN logical judgment expression. The condition part describes one or more process states that need to be satisfied to trigger the constraint atom, and these process states are defined as one or more condition state labels that can be matched with text. The effect part describes the subsequent process impacts after the constraint atom is triggered, and these subsequent process impacts are defined as one or more effect state labels that can be matched with text. Step b, Constructing a causal transmission path: Using the streaming simulation process as a method for executing process design, during the process design, when the effect state label of one constraint atom matches the condition state label of another constraint atom in the streaming simulation, a one-way causal transmission path is established between the two constraint atoms. Step c, execute streaming simulation: input the initial process conditions of the part to be designed to activate the initial constraint atoms that meet the corresponding condition state labels; then, broadcast the effect state label generated by each activated constraint atom as a new process state condition to trigger the matching constraint atoms on all subsequent causal transmission paths in a chain until a set of optimized process parameters, process parameter limits or failure risk warnings are output. In addition, in step a, a weight value characterizing the constraint strength is associated with each constraint atom; and in the streaming derivation of step c, when the triggering conditions of two or more causal transmission paths are simultaneously satisfied and a conflict occurs, the method compares the weight values ​​of the corresponding constraint atoms on each conflicting path, executes the causal transmission path with the highest weight value, and suppresses the remaining causal transmission paths. The method also includes a step of elastically coordinating high-weight conflicts, which includes: when a conflict is detected in two or more causal transmission paths whose weight values ​​are all higher than a system-stored coordinating activation threshold, querying a pre-constructed process element correlation matrix; dynamically generating a temporary constraint atom containing boundary adjustment instructions based on the element with the highest correlation with the conflicting element in the process element correlation matrix; and injecting the temporary constraint atom into the subsequent streaming inference.

2. The rapid design method for aero-engine parts according to claim 1, characterized in that, The process element correlation matrix is ​​constructed by acquiring multidimensional process parameter data from a historical process database and calculating the statistical correlation between any two process parameters X and Y, where the statistical correlation is expressed as a correlation coefficient. The correlation coefficient is calculated according to the following relationship: in This represents the covariance of parameters X and Y. and Let X and Y represent the variances of parameters X and Y, respectively.

3. The rapid design method for aero-engine parts according to claim 1, characterized in that, The method also includes a step of associating constraint atoms with physical execution states and performing real-time tracing. This step includes: when defining the constraint atom, associating the constraint atom that produces a physical effect with a physical signal feature fingerprint extracted from sensor signals corresponding to its normal execution; acquiring physical sensor signals from the manufacturing process in real time and extracting their real-time signal features; when the deviation between the real-time signal features and the physical signal feature fingerprint of any constraint atom exceeds a predetermined range, locking the constraint atom as a potential anomaly source and adjusting its weight value in subsequent streaming inference.

4. The rapid design method for aero-engine parts according to claim 3, characterized in that, The physical signal feature fingerprint is the range of the main peak of the vibration signal in the predetermined frequency band during the processing, and the real-time signal feature is the frequency of the main peak of the spectrum extracted after the vibration signal is acquired in real time and subjected to a fast Fourier transform.

5. The rapid design method for aero-engine parts according to claim 1, characterized in that, The method also includes an adaptive compensation step for raw material batch differences, which includes: comparing the post-production quality inspection data of the first one or more parts in a specific production batch with the expected quality inspection results contained in the constraint atom to calculate one or more batch correction factors, the batch correction factors being used to characterize the material physical properties of the specific production batch; and applying the batch correction factors to dynamically correct the internal key parameters or their weight values ​​of one or more related constraint atoms when performing the flow cytometry simulation on all subsequent parts of the specific production batch.

6. The rapid design method for aero-engine parts according to claim 1, characterized in that, The method also includes a step of performing a systematic health diagnosis of the process system, which includes: continuously recording the conflicts that occur during the streaming simulation and converting the discrete conflict records into continuous time series data characterizing the frequency of conflict occurrence; performing frequency domain analysis on the time series data to identify the periodic patterns exhibited by the conflict occurrence; and generating diagnostic information indicating the existence of systemic risks in the process system based on the periodic patterns.

7. The rapid design method for aero-engine parts according to claim 1, characterized in that, The method also includes a step of actively detecting and adaptively warning of thermal stress abrupt changes during the quenching and cooling process. This step includes: acquiring temperature parameters of two key measuring points, the surface and the core, of the aero-engine part in real time during the quenching and cooling process; calculating a difference characterizing the instantaneous temperature gradient based on the two temperature parameters; and dynamically generating a temporary constraint atom when the difference exceeds a safety threshold related to the part material and stored in the system. The effect of the temporary constraint atom includes an instruction to reduce the flow rate of the cooling medium, and the temporary constraint atom is injected into the causal transmission path of the flow cytometry.

8. The rapid design method for aero-engine parts according to claim 1, characterized in that, In step c, the sequence of all chain-triggered constraint atoms is recorded, and when the failure risk warning is output, the recorded sequence is output as a complete and traceable failure analysis path.

Images