Mes system for automatically adjusting server manufacturing process based on network fault

Abstract

The present application relates to the technical field of fault analysis, in particular to a MES system for automatically adjusting server manufacturing process based on online fault, which comprises a fault data processing module, a fault trend prediction module, a risk component identification module, a propagation path tracking module and a parameter adjustment module.In the present application, through continuous statistical process monitoring on the multi-dimensional quality indicators such as the first power-on failure rate, annualized failure rate and online batch problem of the server, the abnormal fluctuation trend of the quality indicators can be identified in advance before the occurrence of isolated fault events.The system no longer passively responds to the occurred faults, but actively predicts the potential quality risks, and through the fusion analysis of the load, memory usage rate and environmental and operating factors such as temperature and humidity of the data center where the server is located, the complex combination conditions causing the quality indicators to deviate from the normal range are deeply mined, and the specific probability and time window of future fault occurrence are calculated in a quantitative manner.

Description

Technical Field

[0001] This invention relates to the field of fault analysis technology, and in particular to a MES system that automatically adjusts server manufacturing processes based on existing network faults. Background Technology

[0002] The field of fault analysis technology refers to the technical activities of identifying and locating abnormal conditions that occur in servers and related equipment during operation in an information-based manufacturing system. This includes real-time collection of server fault data in the current network operating environment, classification and analysis of fault characteristics, establishment of fault tracing relationships in conjunction with process links, and interaction of analysis conclusions with the manufacturing execution system.

[0003] Among them, the method of adjusting the manufacturing process based on the existing network fault refers to the manufacturing enterprise usually reads and compares the server log file to obtain the fault code or event identifier when the server fails during operation on the existing network, and then modifies or adjusts the manufacturing process in the MES system according to the parameter relationship set in the process standard library.

[0004] Current technologies primarily rely on interpreting server log files after a failure occurs. This approach has significant limitations when dealing with complex or systemic problems. Its core shortcomings lie in the lag in response and the one-sidedness of analysis. This is because the technology must wait for a clear fault code or event identifier to appear before initiating the adjustment process, making it impossible to predict performance degradation or abnormal indicators in their early stages. This leads to missing the best opportunity to prevent problems. Furthermore, it only adjusts based on the preset relationship between a single fault code and process parameters, ignoring the deeper causes that the failure may be triggered by multiple factors coupled together, such as environmental factors, operating load, and specific batches of components. For example, when a batch of servers on the live network experience failures due to a particular brand of memory, current technology may only make minor adjustments to a general motherboard soldering temperature parameter based on the memory error code in the logs. However, this adjustment fails to identify that the root cause of the failure is that the specific brand of memory is more prone to poor soldering under high temperature and high load conditions. This results in a lack of specificity in process adjustments, failing to eradicate hidden dangers, and ultimately causing the same type of failure to recur in different batches of products, resulting in greater quality costs. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a MES system that automatically adjusts server manufacturing processes based on existing network faults.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A MES system for automatically adjusting server manufacturing processes based on existing network faults includes:

[0007] The fault data processing module calculates three indicators for the live network servers: the initial power-on failure rate, the annualized failure rate, and the batch problems on the live network. It then establishes an Xbar-R graph sequence, filters out abnormal indicators in the Xbar-R graph sequence, and generates fault warning instructions.

[0008] The fault trend prediction module analyzes the changes of abnormal indicators corresponding to the fault warning command over time, predicts the probability and time of future faults in the existing network server as the fault trend, and generates fault trend prediction results.

[0009] The risk component identification module identifies future failure trends of existing servers that will negatively impact the overall quality of the existing servers based on the failure trend prediction results, captures the corresponding key components, calculates the comprehensive failure rate of the components, and marks the quality risk components.

[0010] The propagation path tracking module tracks the failure propagation path of the quality risk component in real time, analyzes whether the failure of the quality risk component will affect other components, and generates a failure propagation path diagram.

[0011] The parameter adjustment module automatically adjusts the server manufacturing process parameters and generates a new process configuration table based on the fault trend prediction results of the existing network server and the fault propagation path diagram of the quality risk component.

[0012] As a further aspect of the present invention, the fault warning instruction specifically includes abnormal values ​​of fault indicators, fault occurrence warnings, and fault analysis information; the fault trend prediction results include fault occurrence probability, fault occurrence time, and fault type prediction; the quality risk components specifically include components whose fault frequency exceeds the frequency threshold, fault type-related components, and key performance components; the fault propagation path diagram includes propagation path, impact range, and affected components; and the new process configuration table specifically includes adjusted process parameters, optimized production processes, and improved production flows.

[0013] As a further aspect of the present invention, the fault data processing module includes:

[0014] The historical indicator modeling submodule collects three indicator data of the live network server within a specified period: first power-on failure rate, annualized failure rate, and batch problems in the live network. It establishes an Xbar-R graphic sequence, calculates the center line and upper and lower control limits of the Xbar-R graphic sequence, and establishes historical data control limits for the indicators.

[0015] The live network indicator monitoring submodule obtains the current values ​​of three indicators for the live network server in the current period: the first power-on failure rate, the annualized failure rate, and the live network batch problems. It compares the current value of each indicator with the historical data control limit of the indicator, filters out all indicators that exceed the control limit, and obtains abnormal indicators.

[0016] The fault warning generation submodule, based on the selected abnormal indicators, matches the warning level and handling plan corresponding to each abnormal indicator according to the preset instruction rule library, and generates a fault warning instruction.

[0017] As a further aspect of the present invention, the fault trend prediction module includes:

[0018] The abnormal indicator tracing submodule analyzes the abnormal indicators corresponding to the fault warning command, collects data on the changes of the abnormal indicators over time within a specified period, integrates environmental and operational factors that affect the changes of the indicators, performs trend analysis and qualitative prediction, and establishes a set of factors affecting the changes of the indicators.

[0019] The fault path construction submodule uses the C4.5 decision tree algorithm to calculate the information gain ratio for environmental and operational factors in the set of factors affecting the changes in the indicators. It selects the factor with the highest gain ratio as the decision node and recursively splits the dataset until all branches reach the preset stopping condition. Then, it connects all decision nodes and branches to generate the fault trend path.

[0020] The fault trend quantification submodule locates the decision node at the end of the fault trend path, statistically calculates the occurrence frequency and time distribution of the fault mode represented by the decision node, quantifies it as the probability and time of future faults of the existing network server as the fault trend, and generates fault trend prediction results.

[0021] As a further aspect of the present invention, the integration of environmental and operational factors affecting the changes in indicators includes collecting the temperature and humidity of the computer room where the current network server is located as the environmental factors, and collecting the CPU load rate, memory usage rate, and disk read / write speed of the current network server as the operational factors.

[0022] As a further aspect of the present invention, the step of stopping until all branches reach the preset stopping condition specifically means stopping the segmentation of the current decision node when any of the following conditions are met: all data samples in the current decision node belong to the same fault category, the number of data samples in the current decision node is lower than the minimum number of samples threshold, or the depth of the decision tree reaches the maximum depth threshold.

[0023] As a further aspect of the present invention, the risk component identification module includes:

[0024] The negative trend filtering submodule obtains the future trends of three indicators in the fault trend prediction results: first power-on failure rate, annualized failure rate, and batch problems in the current network. Based on the preset quality risk judgment benchmark, it identifies and filters trends that will have a negative impact on the overall quality of the server, and obtains the future failure trend of the current network server.

[0025] The key component location submodule traces the original fault data associated with the future fault trends of each of the existing network servers, performs secondary qualitative analysis on the original fault data, captures hardware or software that appear together in multiple data records and have an impact on quality as key components, and obtains a list of key components to be evaluated.

[0026] The component risk marking submodule collects the failure records corresponding to the key components in the list of key components to be evaluated, counts the occurrence frequency of key components under each indicator, calculates the overall failure rate of the components, compares it with the risk component threshold, and marks all components that exceed the risk component threshold as quality risk components.

[0027] As a further aspect of the present invention, the propagation path tracing module includes:

[0028] The propagation characteristic calculation submodule collects the deployment location of each quality risk component in the live network server, counts the associated fault records, calculates the frequency and time interval of the impact of each quality risk component failure on other components, quantifies it into the quality risk component propagation probability and propagation delay, and generates component fault propagation data.

[0029] The potential impact analysis submodule dynamically analyzes the potential impact range of a fault based on the potential impact of the quality risk component's propagation probability and propagation delay on other components in the component fault propagation data.

[0030] The propagation path visualization submodule, based on the potential impact range of the fault, takes the quality risk component as the starting node, other affected components as the target nodes, and establishes the fault propagation relationship between each component using the potential impact relationship as the connection path, generating a fault propagation path diagram.

[0031] As a further aspect of the present invention, the parameter adjustment module includes:

[0032] The process correlation analysis submodule calls the fault trend prediction results and the fault propagation path diagram to analyze the causes of failures in quality risk components and the time points when they occur during the manufacturing process, identifies key manufacturing process links that affect server quality, and generates process parameters to be adjusted.

[0033] The parameter adjustment list generation submodule queries the fault phenomenon and process adjustment comparison table in the MES system according to the process parameter items to be adjusted, automatically matches and obtains the corresponding adjustment plan, and generates a process parameter adjustment list.

[0034] The process configuration update submodule updates the existing server manufacturing process parameter settings in the MES system based on the process parameter adjustment list, integrates all the adjusted parameters and formats them to generate a new process configuration table.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0036] In this invention, by continuously monitoring and statistically analyzing multi-dimensional quality indicators such as the server's initial power-on failure rate, annualized failure rate, and batch issues in the current network, abnormal fluctuations in quality indicators can be identified before isolated failure events occur. The system no longer passively responds to existing failures but proactively predicts potential quality risks. By integrating and analyzing environmental and operational factors such as server load, memory usage, and temperature and humidity in the data center, the complex combinations of conditions that cause quality indicators to deviate from normal ranges are deeply explored. The specific probability and time window of future failures are calculated quantitatively, and the key hardware or software causing the problem are accurately located. Furthermore, it reveals how the failure of a single risky component gradually propagates and affects other related components, forming a panoramic understanding of the chain reaction of failures. Finally, this complete analysis, from trend warning to root cause location, is directly linked to specific process parameters in the manufacturing process, achieving automatic, precise, and closed-loop optimization of production process configuration. This fundamentally solves the problem of blind process adjustments caused by broken causal chains. Attached Figure Description

[0037] Figure 1 This is a system flowchart of the present invention;

[0038] Figure 2 This is a flowchart of the fault data processing module of the present invention;

[0039] Figure 3 This is a flowchart of the fault trend prediction module of the present invention;

[0040] Figure 4 This is a flowchart of the risk component identification module of the present invention;

[0041] Figure 5 This is a flowchart of the propagation path tracking module of the present invention;

[0042] Figure 6 This is a flowchart of the parameter adjustment module of the present invention. Detailed Implementation

[0043] 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 and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0044] Please see Figure 1 A MES system for automatically adjusting server manufacturing processes based on existing network faults includes:

[0045] The fault data processing module calculates three indicators for the live network servers: the initial power-on failure rate, the annualized failure rate, and the batch problems on the live network. It then establishes an Xbar-R graph sequence, filters out abnormal indicators in the Xbar-R graph sequence, and generates fault warning commands.

[0046] The fault trend prediction module analyzes the changes in abnormal indicators corresponding to fault warning commands over time, predicts the probability and time of future faults in the existing network servers as fault trends, and generates fault trend prediction results.

[0047] The risk component identification module identifies future failure trends of existing servers that will negatively impact the overall quality of the existing servers based on failure trend prediction results, captures the corresponding key components, calculates the comprehensive failure rate of the components, and marks the components with quality risks.

[0048] The propagation path tracking module tracks the failure propagation path of high-risk components in real time, analyzes whether the failure of high-risk components will affect other components, and generates a failure propagation path diagram.

[0049] The parameter adjustment module automatically adjusts the server manufacturing process parameters and generates a new process configuration table based on the failure trend prediction results of the existing network servers and the failure propagation path diagram of the components with quality risks.

[0050] The fault warning instructions specifically include abnormal values ​​of fault indicators, fault occurrence warnings, and fault analysis information. The fault trend prediction results include fault occurrence probability, fault occurrence time, and fault type prediction. The quality risk components specifically include components whose fault frequency exceeds the frequency threshold, components related to fault type, and key performance components. The fault propagation path diagram includes propagation path, scope of influence, and affected components. The new process configuration table specifically includes adjusted process parameters, optimized production processes, and improved production flows.

[0051] Please see Figure 2 The fault data processing module includes:

[0052] The historical indicator modeling submodule collects three indicator data of the live network server within a specified period: first power-on failure rate, annualized failure rate, and batch problems in the live network. It establishes an Xbar-R graphic sequence, calculates the center line and upper and lower control limits of the Xbar-R graphic sequence, and establishes historical data control limits for the indicators.

[0053] First, data on three indicators—first-power-on failure rate, annualized failure rate, and batch issues in the current network—are collected for a specified server model over several consecutive production cycles. For example, for "Type A" servers, weekly data for these three indicators are collected from week 9 of 2024 to week 8 of 2025, a total of 52 weeks. The first-power-on failure rate refers to the proportion of servers that fail during the initial power-on testing phase after leaving the factory out of all tested servers; the annualized failure rate is the annual failure rate calculated based on the number of failures during a period of server operation; and batch issues in the current network refer to the number of concentrated failure events caused by the same reason within the same batch and model of servers. The collected data is used to establish an Xbar-R control chart sequence for each indicator. In practice, the 52 consecutive weeks of data are divided into 13 subgroups, each containing data from 4 consecutive weeks. Taking the annualized failure rate as an example, the first subgroup contains the annualized failure rate data from week 1 to week 4, which are 0.52%, 0.55%, 0.53%, and 0.56%, respectively. Calculate the mean Xbar1 of the first subgroup, which is (0.52 + 0.55 + 0.53 + 0.56) divided by 4, resulting in 0.54%. Calculate the range R1 of the first subgroup, which is the difference between the maximum value (0.56%) and the minimum value (0.52%) within the subgroup, resulting in 0.04%. Repeat the above calculation process to obtain the means (Xbar1 to Xbar13) and ranges (R1 to R13) of all 13 subgroups. Then, calculate the average of all subgroup means, Xbar-bar, as the center line of the Xbar control chart. Assume the calculated Xbar-bar is 0.58%. Similarly, calculate the average of all subgroup ranges, R-bar, assuming the result is 0.05%. According to the preset control chart coefficient table, for a subgroup with a sample size of 4, the values ​​of coefficients A2, D3, and D4 are 0.729, 0, and 2.282, respectively. These coefficients are standard values ​​obtained by fitting a large amount of experimental data based on statistical process control theory. Their setting is to ensure that the control limits can cover the normal fluctuations of the process at a 99.73% confidence level. The upper control limit UCLx of the Xbar chart is calculated by adding the center line (0.58%), the A2 coefficient (0.729), and multiplying by the R-bar value (0.05%), resulting in 0.61645%. The lower control limit LCLx of the Xbar chart is calculated by subtracting the A2 coefficient (0.729), and multiplying by the R-bar value (0.05%) from the center line (0.58%), resulting in 0.54355%. The center line of the R chart is the R-bar, with a value of 0.05%. The upper control limit UCLr of the R chart is calculated by multiplying the D4 coefficient (2.282) and the R-bar value (0.05%), resulting in 0.1141%. The lower control limit LCLr of the R chart is calculated by multiplying the D3 coefficient (0) and the R-bar value (0.05%), resulting in 0%.Based on the above calculations, the historical data control limits established for the annualized failure rate index are as follows: the center line of the Xbar chart is 0.58%, the upper control limit is 0.61645%, and the lower control limit is 0.54355%; the center line of the R chart is 0.05%, the upper control limit is 0.1141%, and the lower control limit is 0%. The same process was applied to the first power-on failure rate and the batch problems in the existing network, thus establishing the historical data control limits for all indices.

[0054] The live network indicator monitoring submodule obtains the current values ​​of three indicators for the live network server in the current period: first power-on failure rate, annualized failure rate, and live network batch problems. It compares the current value of each indicator with the control limits of historical data for the indicator, filters out all indicators that exceed the control limits, and obtains abnormal indicators.

[0055] First, obtain the current values ​​of various indicators for the "Type A" server in the current period, i.e., week 9 of 2025. Specifically, through the production management system and operation and maintenance monitoring platform, the initial power-on failure rate for this week is collected as 0.45%, the annualized failure rate is 0.63%, and there are 2 batch issues on the live network. After obtaining the current values, compare the current value of each indicator with its corresponding historical data control limits. For the annualized failure rate indicator, compare its current value of 0.63% with the control limits of the Xbar chart established in the previous steps. The upper control limit of the Xbar chart is 0.61645%, and the lower control limit is 0.54355%. Since the current value of 0.63% is greater than the upper control limit of 0.61645%, this indicator is judged to exceed the control limit. Next, compare the initial power-on failure rate indicator. Assume that the upper control limit of the initial power-on failure rate Xbar chart calculated using the same method is 0.50%, and the lower control limit is 0.30%. The current measured initial power-on failure rate is 0.45%, which falls between 0.30% and 0.50%, and therefore does not exceed the control limits. Finally, the indicators for batch problems in the current network are compared. Assuming the upper control limit on the Xbar chart is 1.5 and the lower control limit is 0.2, the current value is 2, which is greater than the upper control limit of 1.5, so this indicator is also judged to exceed the control limits. After comparing all indicators, all indicators that exceed their corresponding historical data control limits are filtered out. In this embodiment, the current values ​​of both the annualized failure rate and the batch problems in the current network exceed their respective upper control limits. Therefore, the final set of abnormal indicators is {annualized failure rate, batch problems in the current network}.

[0056] The fault warning generation submodule, based on the selected abnormal indicators and according to the preset instruction rule library, matches the warning level and handling plan corresponding to each abnormal indicator to generate fault warning instructions.

[0057] Upon receiving the set of abnormal indicators {annualized failure rate, batch issues in the current network}, each abnormal indicator in the set is matched against a pre-defined instruction rule base. This instruction rule base is built based on a retrospective analysis of historical failure data and handling measures, combined with the experience of domain experts. The rule base establishment process includes: statistically analyzing the correlation between the extent to which different indicators exceed control limits and the severity of subsequent quality problems, dividing the exceedance range into different intervals, and assigning a warning level and standardized handling plan to each interval. For example, for the annualized failure rate indicator, the rule base sets: when the indicator value exceeds the upper control limit by 0% to 5%, it is defined as a "Level 3 warning", and the corresponding handling plan is "initiate the root cause analysis process"; when the exceedance is between 5% and 10%, it is defined as a "Level 2 warning", and the handling plan is "notify the production line quality supervisor and suspend the shipment of the relevant batch of products"; when the exceedance exceeds 10%, it is defined as a "Level 1 warning", and the handling plan is "immediately stop the line for inspection and report to the quality management department". The threshold for the early warning level is set by performing regression analysis on 300 historical quality events of different levels to find the optimal dividing point between the abnormal magnitude of the indicator and the level of event impact (quantified and graded based on economic losses and customer impact), ensuring that the false negative rate for high-level early warnings is less than 1% and the false positive rate is less than 5%. In this embodiment, the upper control limit for the annualized failure rate is 0.61645%, the current value is 0.63%, and the excess is (0.63-0.61645) / 0.61645, which is approximately 2.2%. According to the rule base, 2.2% falls within the "0% to 5%" range, therefore the matched early warning level is "Level 3 Early Warning," and the response plan is "Initiate the root cause analysis process." For the batch problem indicator in the current network, the upper control limit is 1.5 events, the current value is 2 events, and the excess is (2-1.5) / 1.5, which is 33.3%. Assuming its rule base defines a rate exceeding 20% ​​as a "Level 1 Warning," this indicator matches a "Level 1 Warning," and the contingency plan is to "establish a cross-departmental problem-solving team to track the problematic batch of materials and equipment." Ultimately, this submodule integrates all matching results and generates two independent fault warning instructions: the first instruction states "Abnormal indicator: Annualized failure rate, Warning level: Level 3, Contingency plan: Initiate root cause analysis process"; the second instruction states "Abnormal indicator: Batch problem in the current network, Warning level: Level 1, Contingency plan: Establish a cross-departmental problem-solving team to track the problematic batch of materials and equipment."

[0058] Please see Figure 3 The fault trend prediction module includes:

[0059] The abnormal indicator tracing submodule analyzes the abnormal indicators corresponding to the fault warning instructions, collects data on the changes of abnormal indicators over time within a specified period, integrates environmental and operational factors that affect the changes of indicators, performs trend analysis and qualitative prediction, and establishes a set of factors affecting indicator changes.

[0060] Among them, the environmental and operational factors that affect the changes in the indicators are integrated, including the temperature and humidity of the computer room where the current network server is located as environmental factors, and the CPU load rate, memory usage rate and disk read / write speed of the current network server as operational factors.

[0061] Upon receiving the fault warning instruction "Abnormal Indicator: Annualized Failure Rate, Warning Level: Level 3", the system begins analyzing the abnormal indicator, the annualized failure rate. First, it collects data points showing the change of this indicator over a specified period, such as the last 8 weeks, forming a time series data set with specific values ​​of {0.58%, 0.57%, 0.59%, 0.60%, 0.61%, 0.61%, 0.62%, 0.63%}. Simultaneously, it integrates environmental and operational factors affecting the indicator's changes. Specifically, the integration of environmental factors involves collecting the weekly average temperature and humidity of the three data centers where the "Type A" servers are primarily deployed over these 8 weeks. For example, the temperature and humidity for week 1 are collected as {22.1℃, 45%RH}, {23.0℃, 46%RH}, and {22.5℃, 44%RH}, respectively, and the average of these readings from the three data centers is calculated as the representative environmental factor value for that week. The integration of operational factors specifically involves collecting weekly average CPU load, weekly average memory utilization, and weekly average disk read / write speeds for servers deployed in the same data center within the same time period. For example, the load, memory utilization, and disk read / write speeds for week 1 were collected as {65%, 70%, 150MB / s}, {68%, 72%, 145MB / s}, and {66%, 69%, 155MB / s}, respectively. The average values ​​were calculated as representative values ​​for the operational factors for that week. These eight weeks of annualized failure rate data, environmental factor data, and operational factor data were mapped one-to-one to form a multidimensional dataset. Trend analysis of this dataset showed that the annualized failure rate gradually increased from 0.58% to 0.63%, exhibiting a continuous, slight upward trend. Furthermore, preliminary qualitative predictions indicate that without intervention, the annualized failure rate is highly likely to remain above 0.63% in the following week. Finally, all collected and integrated data, including the annualized failure rate time series, the data center temperature time series, the data center humidity time series, the CPU load rate time series, the memory usage time series, and the disk read / write rate time series, together constitute the set of factors affecting the changes in indicators.

[0062] The fault path construction submodule uses the C4.5 decision tree algorithm to calculate the information gain ratio for environmental and operational factors in the set of factors affecting the change of indicators. It selects the factor with the highest gain ratio as the decision node and recursively splits the dataset until all branches reach the preset stopping condition. Then, it connects all decision nodes and branches to generate the fault trend path.

[0063] The process continues until all branches meet the preset stopping conditions, specifically, the splitting of the current decision node stops when any of the following conditions are met: all data samples in the current decision node belong to the same fault category, the number of data samples in the current decision node is lower than the minimum number of samples threshold, or the depth of the decision tree reaches the maximum depth threshold.

[0064] The data is processed using a set of factors influencing the changes in indicators. This set of factors influencing the changes in indicators is described in the original text of the preceding module: "All collected and integrated data, including the time series of annualized failure rate, data center temperature, data center humidity, CPU load rate, memory usage, and disk read / write rate, together constitute the set of factors influencing the changes in indicators."

[0065] Using a simplified set of influencing factors containing only four records as an example, each record contains two influencing factors: "Weekly Average Memory Utilization" and "Weekly Average Data Center Temperature," as well as a final classification label. This label is determined based on whether the annualized failure rate for the current week is higher than the historical average of 0.58%, categorized as "High Failure Rate" or "Normal."

[0066] The four records are as follows: 1. Memory usage 70%, server room temperature 22℃, tagged "Normal". 2. Memory usage 75%, server room temperature 26℃, tagged "Normal". 3. Memory usage 85%, server room temperature 23℃, tagged "High Fault". 4. Memory usage 90%, server room temperature 27℃, tagged "High Fault".

[0067] In the C4.5 decision tree algorithm, a node is the basic unit of the decision tree. It includes the root node, which contains the initial complete set of factors affecting the change of indicators; the decision node, which represents the judgment on a certain factor; and the leaf node, which represents the final classification conclusion.

[0068] The optimal split point is determined by calculating the information gain ratio of each candidate influence factor. The information gain ratio is defined by the following complete formula:

[0069] ;

[0070] Among them, information entropy The calculation formula is: ;

[0071] in, This represents the set of factors influencing the changes in the current node's metrics. At the root node where the algorithm begins, it specifically refers to the set of all four records listed above. In subsequent decision nodes, it refers to the subset of records assigned to that node. These are the influencing factors to be evaluated. At the root node, the algorithm will alternately consider "weekly average memory utilization" and "weekly average data center temperature" as influencing factors. Substitute into the formula to evaluate which is the better segmentation criterion. This represents the total number of fault categories. In this scenario, the final classification labels are only "High Fault" and "Normal," therefore... The value of is always 2. This is the index for the fault category. It's a counter variable used to iterate through all faults. Categories. Here. This can represent the "high failure" category. It can represent the "normal" category. The set of influencing factors of indicator changes at the current node In the middle, the first The proportion of each fault category. For example, in the four records of the root node, "high fault" appeared twice, and "normal" also appeared twice, therefore... The value is 2 / 4. The value is also 2 / 4. It is a logarithmic operation with base 2. In information theory, it is used to measure the amount of information. The lower the probability of an event, the greater its information content. The set of factors influencing the changes in indicators at the current node Information entropy. It is used to measure the current set. The "impurity" or "disorder" of a set is determined by its entropy. If all records in the set belong to the same category (e.g., all are "high fault"), the entropy is 0, indicating the purest state. If the number of records in each category is equal, the entropy is the highest (1 in this example), indicating the most disordered state. To use the impact factor The number of subsets generated after dividing the data at a certain split point. For continuous values ​​like "memory usage," the algorithm attempts to find an optimal split point (e.g., >80%) to divide the data in half; the number of subsets generated in this case is... The value is 2. This is the index of the subset. It is a counter variable used to iterate through all subsets. A subset. For example, when splitting by "memory usage > 80%", This can represent a subset of "memory usage ≤ 80%". It can represent a subset of "memory usage > 80%". It is based on the impact factor After partitioning, the resulting number of... A subset. For example, when using "memory usage > 80%" as the splitting criterion, It is a subset that includes records 1 and 2. It is a subset that includes records 3 and 4. The set of factors influencing the changes in indicators at the current node The total number of records contained within. In the root node, The value is 4. For the first Subset The number of records contained therein. Continuing from the previous example, The value is 2, because the subset There are two records; The value is also 2. It is the first Subset Information entropy. It is used to measure the "impurity" within each subset after partitioning. A good partition will significantly reduce the information entropy of the subsets. It is information gain. It represents the numerator of the entire formula. Its physical meaning is: according to the influence factor After the partitioning, how much did the "impurity" of the entire system decrease? Factors with greater information gain have stronger classification capabilities. It is the split information. It represents the denominator of the entire formula. This item is used to measure the impact factor. The "inherent cost" of partitioning. The more subsets a partition produces, and the more fragmented they are, the larger this value becomes. This can be used to penalize factors that take too many values ​​and prevent overfitting. It is the information gain ratio. It is the ratio of information gain to split information, and it is the basis for the C4.5 algorithm to ultimately select the optimal influence factor and split point.

[0072] Step 1: Calculate the initial information entropy Ent(D) of the root node:

[0073] Of the four records, two were marked "high failure" and two were marked "normal". ;

[0074] Step 2: Calculate the information gain ratio of the impact factor "memory utilization" (taking a cutoff point > 80% as an example):

[0075] subset (Memory usage ≤ 80%): {Item 1, Item 2} -> 0 "High Fault", 2 "Normal".

[0076] subset (Memory usage > 80%): {Item 3, Item 4} -> 2 "High Fault", 0 "Normal".

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] .

[0082] Step 3: Calculate the information gain ratio of the influencing factor "computer room temperature" (taking a cutoff point > 25℃ as an example):

[0083] subset (Computer room temperature ≤25℃): {Item 1, Item 3} -> 1 "High Fault", 1 "Normal".

[0084] subset (Computer room temperature > 25℃): {Item 2, Item 4} -> 1 "High Fault", 1 "Normal".

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] .

[0090] Step 4: Select the optimal split point and determine whether to terminate:

[0091] After calculating all candidate influence factors and split points at the root node, the algorithm will compare the results. Based on the aforementioned calculations:

[0092] Information gain ratio segmented by "memory usage > 80%" It is 1.0.

[0093] Information gain rate segmented using "computer room temperature > 25℃" It is 0.

[0094] By comparison, it can be seen that... The value is much greater than This indicates that using "memory utilization > 80%" as the criterion for partitioning can minimize the uncertainty of the system. Therefore, the algorithm chooses "memory utilization > 80%" as the final partitioning criterion for the root node.

[0095] Next, the algorithm examines the state of the new nodes (i.e., subsets) generated by this split. Here, we need to explain the meaning of "completely 'pure'". "Pure" means that all data records within a node belong to the same category label.

[0096] Examine the first new node (subset) generated by the segmentation. (Memory usage ≤ 80%): This node contains {record 1, record 2}, both of which are categorized as "normal". Since there are no outliers (i.e., no "high failure" records) within this node, it is a "completely clean" node.

[0097] Examine the second new node (subset) generated by the segmentation. (Meaning "memory usage > 80%)): This node contains {record 3, record 4}, both of which are categorized as "high failure". Similarly, this node contains no outliers (i.e., no "normal" records), therefore it is also a "completely clean" node.

[0098] Because both new nodes after the split are now in a "completely pure" state, any data reaching these nodes will result in a clear and unambiguous classification conclusion, requiring no further division. In a decision tree, such pure nodes become leaf nodes, representing the endpoint of the decision. Therefore, the algorithm's recursive construction process naturally terminates on these branches. Ultimately, the algorithm successfully constructs a clear and unambiguous failure trend path from the root node to the "high-failure" leaf node: "Memory usage > 80%".

[0099] The fault trend quantification submodule locates the decision node at the end of the fault trend path, statistically calculates the occurrence frequency and time distribution of the fault mode represented by the decision node, quantifies it into the probability and time of future faults of the existing network server as the fault trend, and generates fault trend prediction results.

[0100] Quantitative analysis is performed based on the generated fault trend paths. From these paths, decision nodes at the end of the path are located; these nodes represent specific fault modes. For example, a terminal node might be located where the path is "memory utilization > 80%" and "data center temperature > 25℃". First, the frequency of occurrence of the fault mode represented by this decision node is statistically calculated. Specifically, from the original set of influencing factors of indicator changes, all data records meeting the conditions of "memory utilization > 80%" and "data center temperature > 25℃" are selected. Assume 20 historical records are selected. Then, among these 20 records, the number marked as "high fault" is counted, let's say 18. Therefore, the probability of this fault mode occurring is calculated as 18 divided by 20, which is 90%. Next, the time distribution of these 18 "high fault" records is analyzed. By extracting the timestamp corresponding to each record, the temporal patterns of these fault occurrences are analyzed. For example, it was found that the occurrence times of 15 records were concentrated between 48 and 72 hours after the conditions of "memory utilization > 80%" and "data center temperature > 25℃" were met. Based on this time distribution, the time window for future failures can be quantified. Finally, the probability of occurrence is combined with the occurrence time to quantify the probability and time of future server failures on the current network. The output failure trend prediction result is: when the server's memory utilization is consistently higher than 80% and the data center temperature is higher than 25℃, the probability of the server experiencing an annualized failure rate increase event within the next 48 to 72 hours is 90%.

[0101] Please see Figure 4 The risk component identification module includes:

[0102] The negative trend filtering submodule obtains the future trends of three indicators in the fault trend prediction results: first power-on failure rate, annualized failure rate, and batch problems in the current network. Based on the preset quality risk judgment benchmark, it identifies and filters trends that will have a negative impact on the overall quality of the server, and obtains the future failure trend of the current network server.

[0103] The fault trend prediction result is obtained, namely, "when the server's memory utilization rate is consistently higher than 80% and the temperature of the data center is higher than 25°C, the probability of the server experiencing an annualized failure rate increase event within the next 48 to 72 hours is 90%." This submodule identifies and filters this future trend based on a preset quality risk assessment benchmark. The quality risk assessment benchmark is jointly formulated by industry standards and the company's internal quality red lines. For example, the benchmark is set as follows: any indicator whose predicted failure probability exceeds 70%, or whose prediction leads to the indicator value exceeding the historical control limit, is judged as a trend that will negatively impact the overall quality of the server. This benchmark value is set by retrospectively analyzing 100 major quality incidents in the past, finding that in more than 90% of these incidents, the predicted abnormal probability value of the associated indicators exceeded 70% before the incident occurred. In this embodiment, the predicted occurrence probability is 90%, which is greater than the 70% threshold set in the quality risk assessment benchmark. Therefore, this trend is identified and filtered out as the future fault trend of the existing network server. The final output of the future failure trend of the live network servers is consistent with the aforementioned prediction results, namely, a high probability of an increase in the annualized failure rate caused by high memory usage and high data center temperature.

[0104] The key component location submodule traces the original fault data associated with the future failure trend of each existing network server, performs secondary qualitative analysis on the original fault data, captures hardware or software that appear in multiple data records and have an impact on quality as key components, and obtains a list of key components to be evaluated.

[0105] Upon receiving the future failure trend of the live server, the original failure data associated with this trend is traced. Specifically, the 18 historical data records marked as "high failure" that met the conditions of "memory utilization > 80%" and "data center temperature > 25℃" in the aforementioned steps are traced. A secondary qualitative analysis is performed on the original failure work orders or equipment logs corresponding to these 18 records. The analysis process involves manually or through natural language processing technology extracting detailed descriptions of the failure phenomenon, error codes, and replaced components from each failure record. For example, analyzing record 1 reveals a description of "unexpected device restart, log shows memory ECC error, recovered after replacing B-brand memory module." Analyzing record 2 reveals a description of "system blue screen, dump file points to memory management unit, B-brand memory module replaced." Each of these 18 records is analyzed to identify hardware or software that appears in multiple data records and affects quality. The analysis reveals that 14 of the 18 records explicitly point to a replacement operation due to a "B-brand memory module" failure. In addition, four records mentioned voltage instability alarms for "Model C power supplies," but none of these power supplies were ultimately replaced. By comparison, the frequency of "Brand B memory modules" was significantly higher than other components, and they were the direct cause of the failure. Therefore, "Brand B memory modules" were identified as a critical component, and a list of critical components to be evaluated was obtained. This list currently only contains "Brand B memory modules."

[0106] The component risk marking submodule collects the failure records corresponding to the key components in the list of key components to be evaluated, counts the occurrence frequency of key components under each indicator, calculates the overall failure rate of the component, compares it with the risk component threshold, and marks all components that exceed the risk component threshold as quality risk components.

[0107] Obtain the "Brand B memory module" from the list of key components to be evaluated, and compile its corresponding 14 fault records. To assess its risk level, the component's overall failure rate needs to be calculated, as follows:

[0108] ;

[0109] in, : Represents the annualized failure rate of a component. It is a percentage value that predicts the probability of the component failing after one year of continuous operation, based on data within the observation period. It is a core indicator for assessing its long-term quality risk. This represents the total number of component failures. Within a specified observation period, it is the total number of failure events whose root cause is attributed to "Brand B memory modules" after failure analysis. The unit is "units" or "times". : Represents the total number of hours in a year, which is a constant value. This represents the total operating hours of the component. It refers to the sum of the actual operating hours of all servers equipped with "Brand B memory modules" within the same observation period. This value reflects the cumulative operating time of the entire component group.

[0110] The annualized failure rate calculation process for "Brand B memory modules" is as follows:

[0111] Parameter settings: The setup: By analyzing the root cause of the 18 highly relevant historical fault records identified in step 8, it was ultimately confirmed that the direct cause of 14 faults was a defect in the "Brand B memory module" itself. Therefore, the parameters... Set to 14. Configuration settings: By querying the configuration management database and the device runtime monitoring system, it was found that during the past year's observation period, a total of 2,500 servers on the current network had this "Brand B memory module" installed. The total cumulative runtime of these servers was 17,520,000 hours. Settings: Parameters Represents the total number of hours in a year, i.e. Its value is 8760.

[0112] Substitute the above parameters into the formula for calculating the annualized failure rate:

[0113] ;

[0114] Subsequently, the calculated annualized failure rate of 0.7% was compared with the risk component threshold. The risk component threshold was set based on the historical average failure rate of similar components and supplier quality agreements, specifically set at 1.5 times the historical average annualized failure rate. Statistical analysis of failure data from all brands of memory modules over the past two years revealed a historical average annualized failure rate of 0.15%, therefore the risk component threshold was set at... The 1.5x coefficient is based on a risk management strategy that defines components exceeding the average level by 50% as risk items requiring attention. This strategy has been reviewed and confirmed in conjunction with the supply chain and quality departments. Because the calculated annualized failure rate of 0.7% exceeds the risk component threshold of 0.225%, the "Brand B memory module" was marked as a quality risk component.

[0115] Please see Figure 5 The propagation path tracing module includes:

[0116] The propagation characteristic calculation submodule collects the deployment location of each quality risk component in the live network server, counts the associated fault records, calculates the frequency and time interval of the impact of each quality risk component failure on other components, quantifies it into the quality risk component propagation probability and propagation delay, and generates component fault propagation data.

[0117] An analysis was conducted on the identified quality risk component, "Brand B memory modules." First, the deployment location information of the "Brand B memory modules" in the live network servers was collected. For example, a query through the configuration management database revealed that the memory modules were primarily installed in the DIMM_A1 and DIMM_B1 slots of "Type A" servers. Simultaneously, fault records associated with this memory module were statistically analyzed, specifically the previously identified 14 fault records. A deep analysis of the detailed logs of these 14 records was performed to calculate the frequency and time interval of the impact of "Brand B memory module" failures on other components. The analysis revealed that in 5 of these 14 memory failures, the server's central processing unit (CPU) also recorded a large number of L2 cache correction errors within 24 hours of the failure occurring. This indicates that memory failures may lead to data corruption, thereby affecting the normal operation of the CPU. Therefore, the frequency of the "Brand B memory module" failure impacting the CPU was calculated as 5 divided by 14, approximately 35.7%. Regarding the time interval, the time difference between the occurrence of the memory failure and the first occurrence of the CPU cache error in these 5 related events was calculated, with an average value of 10 hours. These two calculation results are quantified as the probability and delay of the quality risk component propagation. The final component failure propagation data is as follows: the quality risk component "Brand B memory module" has a 35.7% probability of affecting the L2 cache of the "Central Processing Unit" within 10 hours after the failure.

[0118] The potential impact analysis submodule dynamically analyzes the potential impact range of a fault based on the propagation probability and propagation delay of the quality risk component in the component fault propagation data and its potential impact on other components.

[0119] Based on the acquired component failure propagation data—specifically, the 35.7% probability that a quality-risk component, the "Brand B memory module," would affect the CPU's L2 cache within 10 hours of failure—dynamic analysis was conducted. The analysis began with the failure of the "Brand B memory module." Based on the 35.7% propagation probability, the CPU was identified as the direct and highly probable potential target. The scope of the analysis was then expanded. The hardware knowledge base and historical failure association database were consulted to analyze potential subsequent problems caused by CPU L2 cache errors. The analysis showed that persistent L2 cache errors have a high probability (e.g., historical data shows over 60%) of causing operating system kernel crashes, leading to server downtime. Therefore, the potential impact of the failure expanded from the CPU at the hardware level to the stability of the operating system at the system level. Through this layer-by-layer, iterative correlation analysis, a series of chain reactions that could be triggered by a core risk component failure were dynamically depicted, thus determining the potential impact range of the failure: from the "Brand B memory module" failure to the CPU's L2 cache error, and then to the "operating system kernel crash."

[0120] The propagation path visualization submodule, based on the potential impact range of the fault, takes the quality risk component as the starting node and the other affected components as the target nodes, and establishes the fault propagation relationship between each component using the potential impact relationship as the connection path, generating a fault propagation path diagram;

[0121] Based on the potential impact range of the fault, a fault propagation relationship between components is established. The "Brand B memory module," a component at risk of quality issues, is designated as the starting node for fault propagation. Other affected components or system states, namely the "Central Processing Unit" (CPU) and the "Operating System Kernel," are designated as target nodes. The potential impact relationships derived from the analysis are used as connection paths between nodes. Specifically, a directed graph is drawn: a node representing the "Brand B memory module" is created; a node representing the "Central Processing Unit" is created; and a node representing the "Operating System Kernel" is created. A directed edge is drawn from the "Brand B memory module" node to the "Central Processing Unit" node, and propagation characteristic data, such as "propagation probability 35.7%, propagation delay 10 hours," is labeled on this edge. Next, a directed edge is drawn from the "Central Processing Unit" node to the "Operating System Kernel" node, and based on historical data analysis, the corresponding propagation probability is labeled, for example, "propagation probability 60%." By connecting all relevant nodes and paths, a clear fault propagation path diagram illustrating how the fault spreads from one component to another is ultimately generated.

[0122] Please see Figure 6 The parameter adjustment module includes:

[0123] The process correlation analysis submodule calls the fault trend prediction results and fault propagation path diagram to analyze the causes of failures in components with quality risks and the time points when they occur during the manufacturing process, identify key manufacturing process links that affect server quality, and generate process parameters to be adjusted.

[0124] The analysis utilized fault trend prediction results (high memory usage and high-temperature environments leading to increased annualized failure rates) and fault propagation path diagrams (Brand B memory modules being the source of risk). First, a thorough analysis of the causes of failure in "Brand B memory modules" was conducted. Failure analysis (FA) of 14 returned faulty memory modules revealed that the root cause of failure in 12 of them was a cold solder joint on a specific pin of the memory chip. Next, the analysis examined when this cold solder joint issue occurred during the manufacturing process. By tracing the production batch records of this brand of memory modules and comparing them with the server manufacturing process, the surface mount technology (SMT) reflow soldering stage on the server motherboard was identified as a critical step in memory module soldering. Analysis of the reflow soldering process parameters, particularly the temperature profile settings, revealed that at high ambient temperatures, the existing reflow soldering temperature profile might cause uneven heating of some memory chips with larger heat capacities, thus creating the risk of cold solder joints. Therefore, the critical manufacturing process affecting server quality was identified as "motherboard reflow soldering," and the "reflow soldering temperature profile" was identified as a process parameter to be adjusted.

[0125] The parameter adjustment list generation submodule queries the fault phenomenon and process adjustment comparison table in the MES system based on the process parameter items to be adjusted, automatically matches and obtains the corresponding adjustment plan, and generates a process parameter adjustment list.

[0126] Based on the identified process parameter to be adjusted, "Reflow Soldering Temperature Profile," the system queries a pre-defined fault phenomenon and process adjustment comparison table within the Manufacturing Execution System (MES). This comparison table is a knowledge base built upon long-term process experiments and mass production data accumulation. The query criteria are "Fault Phenomenon: Poor Soldering of Memory Chips" and "Associated Component: Brand B Memory Module." The system automatically matches one or more corresponding adjustment plans. For example, one plan might be found that states: "For the poor soldering issue of Brand B memory modules, it is recommended to increase the target temperature of the reflow soldering preheating zone from 150℃ to 155℃ and extend the isothermal zone time by 3 seconds to improve solder joint wettability." After obtaining this specific adjustment plan, the system formats it and generates a process parameter adjustment list. This list clearly shows the equipment number to be adjusted, the parameter name, the current value, and the recommended new value. For example: "Equipment: Reflow oven RF-02, Parameters: Preheating zone temperature, current value: 150℃, adjustment value: 155℃; Parameters: Constant temperature zone time, current value: 60 seconds, adjustment value: 63 seconds."

[0127] The process configuration update submodule updates the existing server manufacturing process parameter settings in the MES system based on the process parameter adjustment list, integrates all the adjusted parameters and formats them to generate a new process configuration table.

[0128] The system receives a list of process parameter adjustments. Based on this list, the operator or automated script updates the existing server manufacturing process parameter settings for the "RF-02 reflow oven" within the Manufacturing Execution System (MES) interface. Specifically, the "preheating zone temperature" parameter is changed from 150℃ to 155℃, and the "isothermal zone time" parameter is changed from 60 seconds to 63 seconds, and the configuration is saved. After the update, the system integrates and formats all adjusted parameters, along with other unchanged process parameters. This formatting process includes recording the parameter modification time, operator, and reason (associated with the fault warning event ID), and organizing them according to the standard process document format. Finally, a new process configuration table with an updated version number is generated. This new process configuration table will serve as the basis for the reflow soldering equipment in subsequent production of this server model.

[0129] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A MES system for automatically adjusting server manufacturing processes based on existing network faults, characterized in that, The system includes: The fault data processing module calculates three indicators for the live network servers: the initial power-on failure rate, the annualized failure rate, and the batch problems on the live network. It then establishes an Xbar-R graph sequence, filters out abnormal indicators in the Xbar-R graph sequence, and generates fault warning instructions. The fault trend prediction module analyzes the changes of abnormal indicators corresponding to the fault warning command over time, predicts the probability and time of future faults in the current network server as the fault trend, and generates fault trend prediction results. The risk component identification module identifies future failure trends of existing servers that will negatively impact the overall quality of the existing servers based on the failure trend prediction results, captures the corresponding key components, calculates the comprehensive failure rate of the components, and marks the quality risk components. The propagation path tracking module tracks the fault propagation path of the quality risk component in real time, analyzes whether the fault of the quality risk component will affect other components, and generates a fault propagation path diagram. The parameter adjustment module automatically adjusts the server manufacturing process parameters and generates a new process configuration table based on the fault trend prediction results of the existing network server and the fault propagation path diagram of the quality risk component. The propagation path tracing module includes: The propagation characteristic calculation submodule collects the deployment location of each quality risk component in the live network server, counts the associated fault records, calculates the frequency and time interval of the impact of each quality risk component failure on other components, quantifies it into the quality risk component propagation probability and propagation delay, and generates component fault propagation data. The potential impact analysis submodule dynamically analyzes the potential impact range of a fault based on the potential impact of the quality risk component's propagation probability and propagation delay on other components in the component fault propagation data. The propagation path visualization submodule, based on the potential impact range of the fault, takes the quality risk component as the starting node, other affected components as the target nodes, and establishes the fault propagation relationship between each component using the potential impact relationship as the connection path, and generates a fault propagation path diagram. The parameter adjustment module includes: The process correlation analysis submodule calls the fault trend prediction results and the fault propagation path diagram to analyze the fault causes of quality risk components and the time points when they occur during the manufacturing process, identify the key manufacturing process links that affect server quality, and generate process parameter items to be adjusted. The parameter adjustment list generation submodule queries the fault phenomenon and process adjustment comparison table in the MES system according to the process parameter items to be adjusted, automatically matches and obtains the corresponding adjustment plan, and generates a process parameter adjustment list. The process configuration update submodule updates the existing server manufacturing process parameter settings in the MES system based on the process parameter adjustment list, integrates all the adjusted parameters and formats them to generate a new process configuration table.

2. The MES system for automatically adjusting server manufacturing processes based on existing network faults as described in claim 1, characterized in that, The fault warning instructions specifically include abnormal values ​​of fault indicators, fault occurrence warnings, and fault analysis information. The fault trend prediction results include fault occurrence probability, fault occurrence time, and fault type prediction. The quality risk components specifically include components whose fault frequency exceeds the frequency threshold, components related to fault type, and key performance components. The fault propagation path diagram includes propagation path, scope of influence, and affected components. The new process configuration table specifically includes adjusted process parameters, optimized production processes, and improved production flows.

3. The MES system for automatically adjusting server manufacturing processes based on existing network faults as described in claim 1, characterized in that, The fault data processing module includes: The historical indicator modeling submodule collects three indicator data of the live network server within a specified period: first power-on failure rate, annualized failure rate, and batch problems in the live network. It establishes an Xbar-R graphic sequence, calculates the center line and upper and lower control limits of the Xbar-R graphic sequence, and establishes historical data control limits for the indicators. The live network indicator monitoring submodule obtains the current values ​​of three indicators for the live network server in the current period: the first power-on failure rate, the annualized failure rate, and the live network batch problems. It compares the current value of each indicator with the historical data control limit of the indicator, filters out all indicators that exceed the control limit, and obtains abnormal indicators. The fault warning generation submodule, based on the selected abnormal indicators, matches the warning level and handling plan corresponding to each abnormal indicator according to the preset instruction rule library, and generates a fault warning instruction.

4. The MES system for automatically adjusting server manufacturing processes based on existing network faults according to claim 3, characterized in that, The fault trend prediction module includes: The abnormal indicator tracing submodule analyzes the abnormal indicators corresponding to the fault warning command, collects data on the changes of the abnormal indicators over time within a specified period, integrates environmental and operational factors that affect the changes of the indicators, performs trend analysis and qualitative prediction, and establishes a set of factors affecting the changes of the indicators. The fault path construction submodule uses the C4.5 decision tree algorithm to calculate the information gain ratio for environmental and operational factors in the set of factors affecting the changes in the indicators. It selects the factor with the highest gain ratio as the decision node and recursively splits the dataset until all branches reach the preset stopping condition. Then, it connects all decision nodes and branches to generate the fault trend path. The fault trend quantification submodule locates the decision node at the end of the fault trend path, statistically calculates the occurrence frequency and time distribution of the fault mode represented by the decision node, quantifies it as the probability and time of future faults of the existing network server as the fault trend, and generates fault trend prediction results.

5. The MES system for automatically adjusting server manufacturing processes based on existing network faults according to claim 4, characterized in that, The integrated environmental and operational factors affecting the changes in the indicators include collecting the temperature and humidity of the data center where the current network server is located as the environmental factors, and collecting the CPU load rate, memory usage rate, and disk read / write speed of the current network server as the operational factors.

6. The MES system for automatically adjusting server manufacturing processes based on existing network faults according to claim 4, characterized in that, The step of stopping until all branches reach the preset stopping condition specifically means stopping the segmentation of the current decision node when any of the following conditions are met: all data samples in the current decision node belong to the same fault category, the number of data samples in the current decision node is lower than the minimum number of samples threshold, or the depth of the decision tree reaches the maximum depth threshold.

7. The MES system for automatically adjusting server manufacturing processes based on existing network faults according to claim 4, characterized in that, For calculating information gain rate The formula used is: ； in, This is the set of factors influencing the changes in indicators at the current node. The influencing factors to be evaluated are: The total number of fault categories. Index for fault categories, The set of factors influencing the change of indicators at the current node In the middle, the first The proportion of each fault category The set of factors influencing the changes in indicators at the current node Information entropy, calculated by the formula is: , To use the impact factor The number of subsets generated after dividing by the split point. The index of the subset. The set of factors influencing the changes in indicators at the current node The total number of records contained therein For the first Subset The number of records contained therein It is information gain. It is split information.

8. The MES system for automatically adjusting server manufacturing processes based on existing network faults according to claim 4, characterized in that, The risk component identification module includes: The negative trend filtering submodule obtains the future trends of three indicators in the fault trend prediction results: first power-on failure rate, annualized failure rate, and batch problems in the current network. Based on the preset quality risk judgment benchmark, it identifies and filters trends that will have a negative impact on the overall quality of the server, and obtains the future failure trend of the current network server. The key component location submodule traces the original fault data associated with the future fault trends of each of the existing network servers, performs secondary qualitative analysis on the original fault data, captures hardware or software that appear together in multiple data records and have an impact on quality as key components, and obtains a list of key components to be evaluated. The component risk marking submodule collects the failure records corresponding to the key components in the list of key components to be evaluated, counts the occurrence frequency of key components under each indicator, calculates the comprehensive failure rate of the components, compares it with the risk component threshold, and marks all components that exceed the risk component threshold as quality risk components.

Images