Blockchain-based smt production data traceability method

Abstract

This invention relates to the field of intelligent manufacturing data management technology and discloses a blockchain-based SMT production data traceability method. The method includes cleaning and standardizing heterogeneous data from the SMT production line, assigning identifiers to batches, and submitting the data for blockchain consensus verification. After verification, on-chain feature extraction is performed to generate data feature vectors containing key process parameters and quality indicators. These vectors are then bound to batch identifiers and stored in a blockchain Merkle tree. When traceability is required, the production health status is analyzed by retrieving the feature vectors of specific batches and performing pattern recognition. Once an anomaly is detected, the depth and scope of traceability are dynamically initiated and precisely customized based on the anomaly pattern characteristics. This method combines the tamper-proof characteristics of blockchain with intelligent data analysis, solving the problems of high data storage pressure and low traceability efficiency in traditional methods, and achieving efficient and accurate intelligent traceability.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing data management technology, specifically a blockchain-based SMT production data traceability method. Background Technology

[0002] In the surface mount technology (SMT) production process, the production line includes various equipment such as pick-and-place machines, solder paste printers, and reflow ovens. These devices generate data with different formats and protocols, forming a complex heterogeneous data stream. Existing technologies typically use centralized databases to store the cleaned and transformed data, or directly upload standardized data to a blockchain for notarization. Both methods aim to solve data silos and trust issues, providing a data foundation for production traceability. However, they are essentially still centered on storing raw data or its simple mapping. Given the high-frequency, multi-parameter characteristics of SMT production, this results in massive data volumes, placing enormous pressure on storage and retrieval.

[0003] A more prominent drawback lies in the lag and rigidity of existing traceability mechanisms. Current traceability methods largely rely on linear retrieval and manual analysis of massive amounts of raw data after the fact. When product quality anomalies occur, quality engineers need to trace back from the final product, reviewing complete historical data records at each stage—a time-consuming, labor-intensive process heavily reliant on human experience. This traceability approach cannot proactively warn of problems at their initial stages, nor can it intelligently assess the potential impact of the anomaly's root cause. This results in a lack of targeted depth and breadth in traceability; either insufficient traceability may miss critical steps, or excessive traceability wastes significant resources. Current technology lacks a mechanism capable of automatically and quickly locating anomaly patterns from massive amounts of data and intelligently driving the traceability process accordingly, making it difficult to meet the efficiency and accuracy requirements of modern intelligent manufacturing. Summary of the Invention

[0004] The purpose of this invention is to provide a blockchain-based method for tracing SMT production data in order to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, this invention provides a blockchain-based SMT production data traceability method, the method comprising: Receive heterogeneous production data streams from multiple SMT production equipment, perform data cleaning and format standardization processing on the heterogeneous production data streams, and generate standardized production data; The standardized production data is divided into production batches, each production batch is assigned a unique identifier, and the production batch data with the identifier is submitted to the blockchain network for distributed consensus verification. In the blockchain network, feature extraction is performed on verified production batch data with identifiers to generate a data feature vector containing key process parameters and quality indicators. The data feature vector is bound to the corresponding production batch identifier and stored in the Merkle tree data structure of the blockchain to form an immutable production data record. Based on real-time production needs, retrieve the data feature vector corresponding to the specific production batch identifier from the blockchain, perform pattern recognition analysis on the data feature vector, and determine the health status of the production process. When an abnormal health status of the production process is identified, a traceability process for the relevant production batch data is initiated based on the feature matching results of the abnormal pattern to determine the depth and scope of the traceability.

[0006] Preferably, the data cleaning and format standardization processing of the heterogeneous production data stream includes: Detect missing and outlier values ​​in data streams from different SMT production equipment, use interpolation algorithms to fill in missing values, and remove outliers that exceed reasonable physical ranges; The completed and cleaned data is converted into a format according to the preset communication protocol specifications to unify the names, types, and units of the data fields; The data after the format is unified is normalized to convert process parameters with different dimensions into a dimensionless numerical range. Add timestamps and data source device identifiers to the normalized data to generate standardized production data blocks with a unified structure.

[0007] Preferably, dividing the standardized production data by production batch includes: Based on the SMT production process, determine the start and end trigger signals for each production batch; Starting from the production batch start trigger signal and ending from the production batch end trigger signal, data segments within the corresponding time interval are extracted from the continuous standardized production data stream. Generate a unique identifier based on a hash algorithm for each extracted data segment. This identifier contains production time, device identifier, and batch number information. Production batch data with unique identifiers is encapsulated into a specific data structure, including a data header and a data body. The data header contains the identifier and metadata, and the data body contains the actual sequence of process parameters.

[0008] Preferably, submitting the identified production batch data to the blockchain network for distributed consensus verification includes: The packaged, identifiable production batch data is broadcast to multiple nodes in the blockchain network; Each node verifies the integrity and legality of the received production batch data, including checking the compliance of the data structure and the validity of the digital signature; Once more than a preset percentage of nodes have passed verification, the production batch data is confirmed as valid data and packaged into the data block to be uploaded to the blockchain. The data blocks are confirmed through a consensus mechanism using proof-of-work. Once confirmed, the blocks are added to the main chain of the blockchain, completing the data on-chain process.

[0009] Preferably, the feature extraction operation performed on the verified, identifiable production batch data in the blockchain network includes: Extract key process parameters from the production batch data that has been put on the chain, including solder paste printing thickness, chip placement accuracy, and reflow soldering temperature profile parameters; Calculate the statistical characteristic values ​​of key process parameters, including mean, variance, extreme values, and trend slope; Quality inspection indicators are extracted from production batch data, including solder joint quality rating, component missing rate, and defective component rate; Statistical feature values ​​and quality inspection indicators are combined into a multidimensional feature vector, and the feature vector is then dimensionality-reduced to retain the most discriminative feature dimensions.

[0010] Preferably, the step of binding the data feature vector with the corresponding production batch identifier and storing it in the Merkle tree data structure of the blockchain includes: Establish a mapping table between data feature vectors and production batch identifiers, and store the mapping table in the blockchain after hashing. The data feature vector itself is compressed and encrypted, then stored in a distributed file system, and the corresponding file hash address is saved in the blockchain; The logic of associating data feature vectors with production batch identifiers is implemented through smart contracts to ensure the integrity and security of data queries. Regularly perform integrity checks on the stored data feature vectors, and verify whether the data has been tampered with by comparing the hash values ​​stored in the blockchain.

[0011] Preferably, retrieving the data feature vector corresponding to a specific production batch identifier from the blockchain includes: Based on the query conditions, the range and time interval of the production batch identifier to be retrieved are analyzed; Locate the block and transaction records containing the target production batch identifier through the blockchain's indexing service; Retrieve the corresponding data feature vector ciphertext from the distributed file system, and decrypt and decompress it using the authorization key; The decrypted data feature vectors are parsed to restore them to an analyzable structured data format.

[0012] Preferably, the pattern recognition analysis of the data feature vector includes: The data feature vectors are input into a trained production process evaluation model, which is trained based on historical normal production data. Calculate the similarity measure between the input data feature vector and the normal production mode feature vector; Based on the magnitude of the similarity metric, the degree of deviation between the current production process and the normal pattern is determined; Multiple threshold ranges are set to classify the health status of the production process into multiple levels based on the degree of deviation.

[0013] Preferably, the process of initiating traceability for relevant production batch data includes: Based on the anomaly level obtained from pattern recognition analysis, the range of historical production batches that need to be traced is determined; Based on the pattern characteristics of the abnormal features, identify the key process steps and quality parameters that require special attention; Tracing the data feature vectors of relevant production batches in reverse chronological order along the production data records stored in the blockchain; By comparing the differences in data feature vectors between abnormal production batches and normal production batches, key process parameters that may lead to the abnormality can be identified.

[0014] Preferably, after determining the depth and scope of the tracing, the method further includes: Based on the identified key process parameters, query the blockchain for other production batch records that use the same process parameters; Analyze the frequency and distribution patterns of abnormal process parameters to identify systemic process problems; The results of the traceability analysis are stored in the blockchain in the form of a structured report, including an assessment of the root cause of the anomaly and the scope of its impact; Based on the traceability analysis results, the process parameter settings of relevant production equipment are automatically adjusted, and the parameter adjustment events are recorded in the blockchain.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By introducing data feature extraction after blockchain consensus verification, key process parameters and quality indicators representing the core characteristics of a production batch are condensed into fixed-dimensional feature vectors, which are then stored as the primary objects in a Merkle tree. This change in technical approach fundamentally alters the data entities stored on the blockchain, transforming them from complex raw records into highly refined data summaries. This reduces the data load on-chain storage and improves the storage scalability and data processing throughput of the blockchain network. Furthermore, because the feature vectors are structured data specifically designed for analysis, the complexity of processing unstructured and semi-structured raw data is avoided during subsequent retrieval, verification, and comparison. This makes reading and analyzing on-chain data extremely efficient, enabling real-time or near-real-time analysis.

[0016] This method uses pattern recognition analysis based on data feature vectors and dynamically determines the depth and scope of traceability based on the identified anomaly patterns. It transforms the traceability decision-making logic from preset, static rules to data-driven, dynamic, and intelligent judgment. Traceability is no longer an indiscriminate scan of all relevant data, but rather a precise and targeted investigation process. The system can automatically identify anomaly characteristics and intelligently infer the most likely source of failure and its potential impact, thus automatically focusing on the most relevant production processes and batches. This avoids wasting computational and human resources on irrelevant data, improving the efficiency and accuracy of traceability operations. It enables the traceability system to possess preliminary proactive analysis and decision-making capabilities, allowing for rapid response to abnormal fluctuations in the production process and realizing a shift from passive querying to proactive intelligent diagnosis. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the blockchain-based SMT production data traceability method described in this invention. Figure 2 A flowchart for cleaning and standardizing heterogeneous production data streams; Figure 3 This is a flowchart of feature extraction for production batch data in a blockchain. Figure 4 This is a histogram of patch precision distribution. Figure 5 This is a bar chart showing the frequency of abnormal process parameters. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a blockchain-based SMT production data traceability method. The method includes: integrating data acquisition interfaces from multiple SMT production equipment to receive heterogeneous production data streams from devices such as pick-and-place machines, printers, and reflow ovens in real time. These data streams may contain different protocol formats and data structures. The received heterogeneous production data streams first enter a data preprocessing module, where data cleaning and format standardization are performed to eliminate noise and inconsistencies, generating standardized production data with a unified structure. The standardized production data is then divided by production batch. Each production batch is defined according to a logical unit or product sequence of the SMT production line, and a unique identifier based on cryptographic hashing is assigned to each batch of data to ensure its uniqueness and traceability. The identified production batch data is submitted to a permissioned blockchain network composed of multiple participating nodes, such as production equipment nodes, quality management nodes, and audit nodes. A distributed consensus mechanism verifies the data to ensure its authenticity and immutability.

[0020] In a blockchain network, verified data blocks trigger smart contracts to perform feature extraction operations. This extracts key process parameters and quality indicators from production batch data, such as solder paste thickness distribution, chip coordinate deviation, and reflow soldering temperature profile characteristics, and combines them into multi-dimensional data feature vectors. These feature vectors are then hash-bound to their corresponding production batch identifiers and stored in the blockchain's Merkle tree data structure. The root hash of the Merkle tree ensures the integrity of the entire data record. When anomalies occur during production or an audit is required, the system retrieves the data feature vectors associated with specific batch identifiers from the blockchain based on real-time query needs. Pattern recognition algorithms are used to analyze the vector distribution and patterns to assess the health of the production process. If an abnormal pattern is identified, such as parameter deviation or quality degradation, the system automatically initiates a traceability process. Based on the matching results of the abnormal features, the system determines the depth of the traceability, such as tracing back to previous batches or related process stages, thereby pinpointing the root cause.

[0021] Example 1: See Figure 2The data cleaning and format standardization stages are implemented through a data acquisition gateway deployed in the production environment. This gateway continuously monitors heterogeneous data streams from various devices on the surface mount technology production line. These heterogeneous data streams include the position coordinates of the pick-and-place machine, pressure readings from the solder paste printer, temperature sensor data from the reflow oven, and image analysis results from the automated optical inspection system. The data cleaning module first performs missing value detection on the heterogeneous data streams. The missing value detection algorithm identifies the intervals between data points based on time series analysis. When a missing data point is detected, a linear interpolation algorithm is used to fill in the missing value. The linear interpolation algorithm calculates the intermediate value based on the preceding and following valid data points. Simultaneously, the outlier detection module runs statistical process control rules to identify values ​​that exceed reasonable physical ranges, such as temperature readings exceeding 300 degrees Celsius or negative values. These outliers are directly discarded to ensure data quality. In practice, the completed and cleaned data enters the format conversion stage. The format conversion module operates according to the preset communication protocol specifications, mapping data fields from different devices to a unified data model. For example, it converts register values ​​in the Modbus protocol to floating-point numbers, renames OPCUA node data to standard field names, and standardizes units, such as converting Fahrenheit to Celsius and timestamps to ISO8601 format. The format-unified data then undergoes normalization processing. This normalization process uses a minimum-maximum scaling method to linearly transform the original data to the [0,1] interval, eliminating the influence of different dimensions. For example, voltage signals are converted from 0-5V to dimensionless values ​​between 0 and 1. After normalization, precise timestamps and data source device identifiers are added to the data blocks. The timestamps are taken from a high-precision clock source, and the data source device identifiers include the device serial number and location code. This generates standardized production data blocks with a unified structure. The standardized production data blocks contain a header and a body; the header stores metadata, and the body stores the process parameter sequence.

[0022] In some embodiments, standardized production data is divided into production batches. The production batch division logic is based on the surface mount technology (SMT) production process definition. The system configures batch start and end trigger signals. The start trigger signal may come from the solder paste printer's start event, and the end trigger signal may come from the reflow oven's completion signal. From the continuous standardized production data stream, data segments within corresponding time intervals are extracted, starting from the start trigger signal and ending at the end trigger signal. The extraction operation ensures that the data segment completely covers all activities of a production batch, and the data segment contains the entire process sequence from printing to soldering. A unique identifier based on the SHA-256 hash algorithm is generated for each extracted data segment. The generation process of the unique identifier connects the production timestamp, device identifier, and batch sequence information to generate a hash value as the unique identifier. The unique identifier guarantees global uniqueness and collision prevention. In specific implementations, after generating the unique identifier, the data segment is encapsulated into a specific data structure. The specific data structure includes a data header and a data body. The data header contains the unique identifier, metadata such as version number and cryptographic digest, and the data body contains the actual process parameter sequence, such as a list of placement coordinates or temperature profile points. Understandably, the data encapsulation process ensures that the data structure is compact and parsable, facilitating subsequent blockchain submissions.

[0023] Optionally, the outlier removal module in data cleaning and format standardization can employ various statistical methods. In some embodiments, outlier detection uses the Z-score algorithm, calculating the Z-score value for each data point. When the absolute Z-score exceeds 3, it is considered an outlier and removed. In specific implementations, spline interpolation can also be used to complete missing values. Spline interpolation, based on polynomial fitting, provides a smoother data completion effect and is suitable for non-linearly changing production parameters. The communication protocol specification in the format conversion stage may support multiple industry standards, such as converting a custom binary protocol to JSON format, ensuring complete consistency in the names, types, and units of data fields. The minimum-maximum scaling method in normalization requires predefining the theoretical range of each parameter, obtained from the equipment specifications. For example, the solder paste thickness range is set to 50-150 micrometers, which is then scaled to the 0-1 range. When adding timestamps and data source device identifiers, the timestamps are synchronized using the Network Time Protocol (NTP), and the data source device identifier includes the MAC address or IP address, enhancing traceability accuracy.

[0024] In some embodiments, the trigger signal for production batch segmentation can be flexibly configured. The start trigger signal may be based on a change in the product serial number, and the end trigger signal may be based on a quality inspection completion event, ensuring that batch segmentation is consistent with production logic. The time interval calculation for data segment extraction takes into account device latency, and the system incorporates buffer time to prevent data loss. Extracted data segments are temporarily stored in a cache. Unique identifier generation uses salted hashing to enhance security; the salt value comes from a random number generator to prevent hash collisions. The data encapsulation structure design supports expansion; the data header can include a compression flag or an encryption algorithm identifier, and the data body uses a binary format to reduce storage space.

[0025] Optionally, the missing value detection in the data cleaning module can integrate a machine learning model. This model predicts missing values ​​based on historical data, improving completion accuracy. In practice, the format conversion module may include data validation rules. These rules check field integrity, such as whether required fields are empty; invalid data is rejected and logged. Normalized data blocks are given digital signatures using asymmetric encryption algorithms to ensure the authenticity of the data source. After generating unique identifiers for each production batch, the system records an identifier mapping table, which is stored in a local database for quick retrieval. The data structure in the data encapsulation stage may use the TLV format, which supports flexible expansion of new parameters without affecting backward compatibility.

[0026] It is understandable that the data stream processing pipeline for data cleaning and format standardization is designed with a parallel architecture. This parallel architecture allows for the simultaneous processing of data from multiple devices, improving throughput. In some embodiments, outlier-removed data undergoes smoothing filtering using a moving average algorithm to reduce the impact of random noise. Unit unification during the format conversion stage may involve complex conversions, such as converting pressure units from psi to kPa, ensuring global consistency. Normalization parameters are dynamically adjusted based on real-time data statistics to adapt to production changes. High-resolution timers are used when adding timestamps, achieving millisecond-level accuracy to meet high-precision traceability requirements. The production batch segmentation algorithm optimizes memory usage, streaming processing avoids large data volume blocking, and unique identifiers are submitted to the next stage immediately after generation, reducing latency. The data header encapsulation includes a checksum for integrity verification, preventing transmission corruption.

[0027] Example 2: See Figure 3After the production batch data with identifiers is generated by the data encapsulation module, it is transmitted to the blockchain client interface. The blockchain client interface serializes the data into transaction data in a specific format. The transaction data includes the hash value of the production batch data, the submitter's digital signature, and a timestamp. The serialized transaction data is broadcast to multiple nodes in the blockchain network. Node types include production workshop server nodes, quality management center nodes, and third-party auditing agency nodes. Each node runs the same blockchain client software and maintains a complete copy of the ledger. After receiving the broadcast transaction data, the nodes first verify whether the structure of the transaction data conforms to the predefined specifications, check the integrity of the fields, the data format, and the validity of the digital signature. Digital signature verification uses the submitter's public key to decrypt the signature digest and compare it with the hash value calculated from the transaction data to ensure that the data source is authentic and has not been tampered with.

[0028] In some embodiments, nodes further verify the legality of production batch data. This legality verification is accomplished by executing business rules encoded in a smart contract. The smart contract code defines the logic for checking parameter reasonableness, such as whether the solder paste printing thickness is within the equipment's allowable range or whether the pick-and-place machine coordinates exceed the physical working area. The verification process also includes checking the uniqueness of the production batch identifier to prevent duplicate submissions of the same batch of data and ensure that each record on the blockchain corresponds to a unique production activity. When more than half of the nodes in the network independently complete the verification and reach a consensus, the transaction is marked as a valid transaction, and valid transactions are collected into the node's transaction pool to await packaging. The consensus node selects a batch of valid transactions from the transaction pool and combines them with the hash value of the previous block, the current timestamp, and a random number to construct a new candidate block. The consensus mechanism uses proof-of-work, where nodes need to solve a cryptographic puzzle: finding a random number such that the hash value of the candidate block meets a difficulty target, such as the first few bits of the hash value must be zero.

[0029] Understandably, the proof-of-work process requires significant computational resources. The first node to successfully find a valid random number broadcasts the new block to the entire network. Other nodes, upon receiving the new block, immediately verify its validity. Verification includes the legality of all transactions within the block, the correctness of the proof-of-work, and whether the block hash value meets the difficulty requirements. After successful verification, the node appends the new block to its local blockchain main chain, completing the data upload process. The production batch data thus becomes an immutable and permanent record. In practice, the feature extraction operation is triggered by a dedicated smart contract deployed in the blockchain network. When the smart contract detects the successful upload of new production batch data, it automatically calls the feature extraction function. The feature extraction function accesses the production batch data stored on the blockchain, first parsing the process parameter sequence in the data body to extract key process parameters. The selection of key process parameters is based on expertise in surface mount technology, such as extracting key points on the temperature curve from reflow oven data, including the initial heating temperature, peak temperature, time above the liquidus line, and cooling rate.

[0030] In practical implementation, solder paste printing thickness parameters are extracted from the printer's sensor data, calculating the average and standard deviation of multiple measurement points within a printing cycle. Placement accuracy parameters are obtained from the placement machine's coordinate data, analyzing the mean and extreme values ​​of the deviation between the component placement position and the target position. The statistical feature calculation module performs calculations on the extracted key process parameters, generating statistical feature values ​​such as mean, variance, and trend slope. For example, the average temperature rise slope of the reflow soldering temperature curve in the 150°C to 200°C range is calculated using the following formula: ; in: Represents the average temperature rise slope. This represents the number of temperature sampling points. and This represents the temperature values ​​of two adjacent sampling points. and This represents the corresponding sampling timestamp. Meanwhile, quality inspection indicators are extracted from data from automated optical inspection systems or online testing instruments. These indicators include solder joint yield, number of missing components, and number of defective components, and are quantified into numerical forms. The extracted statistical feature values ​​and quality inspection indicators are combined into a multidimensional feature vector. Since the multidimensional feature vector may have high dimensionality, dimensionality reduction is performed using principal component analysis (PCA). PCA identifies the direction of maximum variance in the data, projects the original features onto a few principal components, preserves the main variation information of the data, and generates the final low-dimensional data feature vector for subsequent analysis.

[0031] In some embodiments, the node verification logic of distributed consensus verification can include more complex rules, such as checking whether the timestamps of production batch data are consistent with equipment logs to prevent data backfilling or prefilling. In specific implementations, the difficulty target of proof-of-work can be dynamically adjusted according to network computing power, and the dynamic adjustment mechanism ensures that the block generation rate remains stable. The list of key process parameters in feature extraction operations can be dynamically configured to adapt to new production processes or equipment by updating smart contract code. It can be understood that, in addition to principal component analysis, linear discriminant analysis can also be used for dimensionality reduction. Linear discriminant analysis considers category label information while reducing dimensionality, improving the discriminative power of feature vectors.

[0032] Optionally, the blockchain client interface can encrypt transaction data before broadcasting it. Encryption uses the public key of the receiving node to ensure that only authorized nodes can decrypt and view the data content. In practice, the execution result of the feature extraction smart contract, i.e., the data feature vector, can be selectively stored back on the blockchain or written to an off-chain database, and its hash value is stored on the blockchain as evidence. The principal component analysis model for feature vector dimensionality reduction can be pre-trained based on historical data, with model parameters stored in the smart contract or obtained from off-chain via an oracle service.

[0033] See Figure 4 This chart, a histogram, is a core data visualization tool used in the feature extraction phase. It focuses on the frequency distribution of placement accuracy, a key process parameter for SMT (Surface Mount Technology) machines. The horizontal axis represents the numerical range of placement accuracy, and the vertical axis represents the frequency of occurrence within each range. Different colored bars distinguish the distribution of different accuracy ranges. This chart represents the preliminary analysis results of feature extraction after production batch data is uploaded to the blockchain: placement accuracy is one of the key process parameters. This histogram visually presents its numerical distribution characteristics, providing a foundation for subsequent calculations of the parameter's mean, variance, and other statistical characteristics, ultimately supporting the generation of multi-dimensional data feature vectors. Simultaneously, this distribution reflects the operational stability of the placement machine, serving as crucial preliminary data support for subsequent pattern recognition and anomaly tracing, aligning with the technical logic of extracting core features from on-chain data to improve traceability efficiency.

[0034] Example 3: The binding process between data feature vectors and production batch identifiers begins with the establishment of a mapping table. This mapping table is a data structure containing corresponding entries for the hash values ​​of data feature vectors and production batch identifiers. It is stored in key-value pairs, where the key is the production batch identifier and the value is the cryptographic hash value of the data feature vector. After generation, the mapping table undergoes hashing, for example, using the SHA-256 algorithm to calculate the root hash of the entire table. This root hash is stored in a blockchain transaction record as immutable evidence of the mapping relationship. Simultaneously, due to the large data volume of the data feature vectors, they undergo compression and encryption. Compression uses the LZ77 algorithm to reduce data volume, and encryption uses the AES-256 algorithm combined with a symmetric key. The encrypted data feature vector is stored in a distributed file system, such as IPFS or the Hadoop Distributed File System. The distributed file system returns a content identifier or file hash address, a unique string pointing to the storage location. This file hash address is recorded in another blockchain transaction and associated with the root hash of the mapping table, ensuring the coordination between on-chain evidence storage and off-chain storage.

[0035] In some embodiments, a smart contract is deployed to a blockchain network to implement the logic for associating data feature vectors with production batch identifiers. The smart contract code is written in Solidity or Chaincode and includes a query function. This function accepts the production batch identifier as input, retrieves the file hash address of the data feature vector by looking up the on-chain record corresponding to the hash value in the mapping table, and then sends a request to the distributed file system to retrieve the ciphertext. The smart contract also includes access control logic to verify the querier's digital signature and permission level, ensuring that only authorized users can retrieve the data feature vector. A periodic integrity verification mechanism is implemented through a scheduled task. This task runs at fixed time intervals, such as once a day. The verification process downloads the ciphertext of the data feature vector from the distributed file system, reverse-calculates the current hash value using the same compression and encryption algorithms, and compares the current hash value with the file hash address stored in the blockchain. If the two hash values ​​match, the data is intact; otherwise, an alarm is triggered to record the abnormal event. The integrity verification formula is expressed as: ; in: This represents the hash address value of the file stored in the blockchain. This represents the hash value calculated in real time after retrieving the encrypted data feature vector from the distributed file system. The verification result is a boolean value indicating the data status. In specific implementations, when retrieving the data feature vector corresponding to a specific production batch identifier from the blockchain, the system first parses the query conditions submitted by the user. The query conditions may include a list of production batch identifiers, a time range, or a device identifier. The parsing module converts the query conditions into a standardized format and then calls the blockchain's indexing service. The indexing service uses a B-tree index built based on the block height or transaction hash to quickly locate the block and transaction record containing the target production batch identifier. The location process is achieved by traversing the blockchain's header information to match timestamps and identifier patterns.

[0036] Optionally, after the indexing service returns the transaction record, the system extracts the file hash address from the record and then sends a retrieval request to the distributed file system. This request includes the file hash address and an authentication token. After verifying the token's validity, the distributed file system returns the encrypted data feature vector. This encrypted data is decrypted using an authorization key obtained from the security key management service. After decryption, the data is decompressed and restored to its original binary stream. The decompression process uses the reverse of the LZ77 algorithm to restore the original size of the data feature vector. The decrypted and decompressed data feature vector undergoes format parsing. The format parsing module converts the binary stream into a structured data format, such as an array or dictionary, based on a predefined pattern description. This structured data format contains feature dimensions and values, facilitating subsequent analysis by tools.

[0037] In some embodiments, the hash operation of the mapping table may be optimized using a Merkle tree structure. Multiple mapping table entries are organized into a Merkle tree, with leaf nodes representing the hash of a single entry, and non-leaf nodes representing the merged hash of their child nodes. The root hash is stored in the blockchain, improving query efficiency and verifiability. In specific implementations, the compression and encryption process of data feature vectors can be processed in blocks. Large feature vectors are divided into fixed-size blocks, each of which is independently compressed and encrypted. The block hash chain ensures sequential integrity. The query logic of smart contracts may support batch retrieval, allowing queries of multiple production batch identifiers at once, reducing the number of blockchain interactions. The scheduling of periodic integrity checks can be event-triggered, such as automatic verification after each data access, enhancing real-time performance.

[0038] Optionally, the choice of distributed file system may consider performance requirements, such as using IPFS for high-concurrency reads or using private distributed storage for low-latency scenarios. In specific implementations, the format parsing of data feature vectors may include type checking to ensure that feature values, such as floating-point numbers or integers, are correctly converted to avoid parsing errors. Integrity verification alarm mechanisms can be integrated into the monitoring system to automatically notify administrators to handle data anomalies. Authorization key management during the retrieval process is enhanced with hardware security modules to prevent key leakage. The process in this embodiment is implemented through modular design; components such as the parsing module, indexing service, and decryption module communicate through API interfaces, ensuring system modularity and maintainability.

[0039] Example 4: The process of pattern recognition analysis using data feature vectors begins by inputting the data feature vectors into a trained production process evaluation model. This model is a machine learning model trained on historical normal production data. The model type can be a support vector machine or a deep neural network. The model receives multi-dimensional data feature vectors as input, which include statistical feature values ​​of key process parameters and quality inspection indicators. The production process evaluation model calculates the similarity measure between the input data feature vectors and the normal production pattern feature vectors stored within the model. The similarity measure uses the cosine similarity algorithm, which evaluates similarity by calculating the cosine of the angle between two vectors in space. The formula is expressed as: ; in: Represents the cosine similarity value. This represents the feature vector of the current input data. Represents the feature vector of normal production mode. Dot product operation representing vectors, and These represent the magnitudes of the two vectors, respectively. Similarity metric. The similarity score ranges from -1 to 1, with values ​​closer to 1 indicating higher similarity. Based on the calculated similarity score, the system determines the degree of deviation between the current production process and the normal pattern. The degree of deviation is defined as... The larger the deviation value, the higher the probability of an anomaly.

[0040] In some embodiments, the system sets multiple threshold intervals to classify the health status levels of the production process. These multiple threshold intervals are predefined based on historical data analysis, mapping the degree of deviation to discrete status levels. Refer to Table 1 for the correspondence between threshold intervals and health status levels.

[0041] Table 1: Classification of Health Status Levels in Production Processes

[0042] When pattern recognition analysis identifies an anomaly in the production process health status, specifically a health status level of 3 or 4, the system automatically initiates a traceability process for the relevant production batch data. The traceability process first determines the range of historical production batches to be traced based on the anomaly level obtained from pattern recognition analysis. For example, a level 3 anomaly might trace the current batch and the previous five batches, while a level 4 anomaly might trace the current batch and the previous twenty batches. After determining the traceability range, the system analyzes the contribution of each dimension in the data feature vector based on the pattern characteristics of the anomaly, identifying the key process steps and quality parameters requiring focus. For instance, if the reflow soldering temperature-related feature dimension deviates the most, the traceability process will focus on examining the historical data of the reflow oven equipment.

[0043] In practice, the data feature vectors of relevant production batches are traced backwards in chronological order along the production data records stored in the blockchain. Starting with the batch where an anomaly was identified, the data feature vector records of previous batches stored in the blockchain are retrieved sequentially. The retrieval process is implemented by calling the query interface of the blockchain smart contract. The smart contract returns historical data feature vectors in chronological order based on the timestamp of the production batch identifier. By comparing the differences between the data feature vectors of the abnormal production batch and the normal production batch, the offset of each feature dimension is calculated through difference analysis to locate the key process parameters that may lead to the anomaly. The location of key process parameters is accomplished by comparing the numerical differences between the abnormal batch and the normal batch in a specific dimension. Parameters with differences exceeding a preset threshold are marked as suspicious parameters.

[0044] Optionally, the production process assessment model in pattern recognition analysis can be updated periodically, retraining the model parameters using the latest normal production data to adapt to slow changes in production processes. In specific implementation, similarity measurement calculations can consider the weighting of feature vectors, assigning different weights to different process parameters, with important parameters such as solder joint quality-related features receiving higher weights. The threshold range for health status level classification can be dynamically adjusted according to specific product models, with higher-precision products using stricter thresholds. The batch range determination for the traceability process can be combined with production cycle time, converting the time window into specific batch numbers. It can be understood that the pattern analysis of abnormal features can employ clustering algorithms to identify the similarity between the distribution pattern of abnormal feature vectors and historical abnormal cases.

[0045] In some embodiments, the reverse tracing process can optimize retrieval performance by establishing a cached index for blockchain data to reduce on-chain query latency. In specific implementations, the comparison of data feature vector differences can be aided by visualization tools, generating feature dimension difference charts for easy manual assessment. The location results of key process parameters can be fed back to the production process evaluation model as input for feature importance analysis. Optionally, after the traceability process is initiated, a traceability task log can be generated, recording the batch range being traced, the process parameters of interest, and the analysis results. The log is stored on the blockchain for auditing. In specific implementations, the determination of key process steps can be based on an expert knowledge base containing mapping rules between different anomaly patterns and process steps. Preset thresholds in the difference analysis can be dynamically calculated based on the process capability index to improve location accuracy.

[0046] Example 5: The system queries other production batch records using the same process parameters in the blockchain. The query process is implemented by calling the query function of the blockchain smart contract. The smart contract receives the identifier of the key process parameter and the allowed deviation range as input parameters, traverses the production batch data records stored in the blockchain, and filters out all historical production batches whose process parameter values ​​fall within the specified range. The query results return a list of matching production batch identifiers and their corresponding process parameter values. The system organizes these records into a dataset for subsequent analysis, analyzing the frequency and distribution patterns of abnormal process parameters, calculating the percentage of each process parameter value that appears in the historical data out of the total number of batches, and identifying systemic process problems. The formula for calculating the frequency of occurrence is expressed as: ; in: This represents the frequency of occurrence of a specific process parameter value. This represents the number of batches in the historical production batches that have this parameter value. This represents the total number of production batches within the query range. It also indicates the frequency of occurrence of a specific process parameter value. When the preset threshold is exceeded, the system determines that there is a systemic process problem and the root cause needs to be further investigated.

[0047] In some embodiments, the traceability analysis results are stored in the blockchain in the form of a structured report. This structured report contains multiple standardized fields, including root cause analysis, impact scope assessment, recommended actions, and a report generation timestamp. Root cause analysis is based on the frequency and distribution patterns of process parameters, combined with equipment maintenance records and material batch information for correlation analysis. The impact scope assessment statistically analyzes the number of production batches affected by the anomaly and their quality grade distribution. After being digitally signed, the structured report is stored in a distributed ledger via blockchain transactions, ensuring the report's immutability and auditability. Once stored, the system automatically adjusts the process parameter settings of relevant production equipment based on the traceability analysis results. The adjustment logic is determined according to the anomaly type and root cause analysis results. For example, when a reflow oven temperature deviation is identified as causing poor solder joints, the system sends a parameter adjustment command through the equipment control interface to correct the temperature setpoint to the normal range.

[0048] In practice, after the parameter adjustment command is sent to the production equipment controller, the system records the parameter adjustment event in the blockchain. The event record includes the parameter value before adjustment, the parameter value after adjustment, the adjustment time, the executing equipment identifier, and a description of the adjustment reason. The event record is submitted to the blockchain network as a new transaction, and after distributed consensus verification, it is added to the blockchain, forming a complete parameter adjustment history. The entire process achieves closed-loop control from problem identification to corrective measures, ensuring continuous improvement of the production process. After parameter adjustment, the system continues to monitor the changing trends of relevant process parameters, verify the effectiveness of the adjustment measures, collect monitoring data in real time, and compare it with the data before adjustment to confirm whether the production process has returned to normal.

[0049] Optionally, when querying other production batch records using the same process parameters, analysis can be performed segmented by time dimension to statistically analyze parameter distribution patterns within different time periods and identify long-term drift trends in process parameters. In practical implementation, the identification of systemic process problems can be combined with statistical process control methods, using control charts to monitor the stability of process parameters and triggering alarms when parameter points exceed control limits. The impact assessment of structured reports can include impact analysis on downstream processes, assessing the potential impact of abnormal process parameters on the quality of the final product. When automatically adjusting process parameter settings, a gradual adjustment strategy can be adopted to avoid production fluctuations caused by sudden parameter changes, adjusting small amounts each time and observing the effects.

[0050] See Figure 5This chart, a core visualization for the anomaly tracing and process optimization phase, is a horizontal bar chart focusing on the frequency distribution of five types of abnormal process parameters in SMT production. The vertical axis represents the type of abnormal process parameter, and the horizontal axis represents the frequency of the parameter in historical production batches. Different colored bars distinguish different parameters. This chart was compiled after the system queried historical batch records with the same parameters in the blockchain, based on the identified key process parameters of the abnormal production batch. The chart shows that the frequency of solder joint quality is significantly higher than other parameters, clearly indicating that it is the most prominent systemic process risk point in current production. This result supports the core objective of identifying systemic process problems and provides data for subsequent generation of structured traceability reports and automatic adjustment of equipment process parameters. It aligns with the technical logic of closed-loop control from anomaly tracing to process optimization and is a key intermediate analysis result for achieving continuous improvement in the production process.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A blockchain-based method for tracing SMT production data, characterized in that, Includes the following steps: Receive heterogeneous production data streams from multiple SMT production equipment, perform data cleaning and format standardization processing on the heterogeneous production data streams, and generate standardized production data; The standardized production data is divided into production batches, each production batch is assigned a unique identifier, and the production batch data with the identifier is submitted to the blockchain network for distributed consensus verification. In the blockchain network, feature extraction is performed on verified production batch data with identifiers to generate a data feature vector containing key process parameters and quality indicators. The data feature vector is bound to the corresponding production batch identifier and stored in the Merkle tree data structure of the blockchain to form an immutable production data record. Based on real-time production needs, retrieve the data feature vector corresponding to the specific production batch identifier from the blockchain, perform pattern recognition analysis on the data feature vector, and determine the health status of the production process. When an abnormal health status of the production process is identified, a traceability process for the relevant production batch data is initiated based on the feature matching results of the abnormal pattern to determine the depth and scope of the traceability.

2. The blockchain-based SMT production data traceability method according to claim 1, characterized in that, The data cleaning and format standardization process for the heterogeneous production data stream includes: Detect missing and outlier values ​​in data streams from different SMT production equipment, use interpolation algorithms to fill in missing values, and remove outliers that exceed reasonable physical ranges; The completed and cleaned data is converted into a format according to the preset communication protocol specifications to unify the names, types, and units of the data fields; The data after the format is unified is normalized to convert process parameters with different dimensions into a dimensionless numerical range. Add timestamps and data source device identifiers to the normalized data to generate standardized production data blocks with a unified structure.

3. The blockchain-based SMT production data traceability method according to claim 2, characterized in that, The step of dividing the standardized production data by production batch includes: Based on the SMT production process, determine the start and end trigger signals for each production batch; Starting from the production batch start trigger signal and ending from the production batch end trigger signal, data segments within the corresponding time interval are extracted from the continuous standardized production data stream. Generate a unique identifier based on a hash algorithm for each extracted data segment. This identifier contains production time, device identifier, and batch number information. Production batch data with unique identifiers is encapsulated into a specific data structure, including a data header and a data body. The data header contains the identifier and metadata, and the data body contains the actual sequence of process parameters.

4. The blockchain-based SMT production data traceability method according to claim 3, characterized in that, The step of submitting the identified production batch data to the blockchain network for distributed consensus verification includes: The packaged, identifiable production batch data is broadcast to multiple nodes in the blockchain network; Each node verifies the integrity and legality of the received production batch data, including checking the compliance of the data structure and the validity of the digital signature; Once more than a preset percentage of nodes have passed verification, the production batch data is confirmed as valid data and packaged into the data block to be uploaded to the blockchain. The data blocks are confirmed through a consensus mechanism using proof-of-work. Once confirmed, the blocks are added to the main chain of the blockchain, completing the data on-chain process.

5. The blockchain-based SMT production data traceability method according to claim 4, characterized in that, The feature extraction operation performed on the verified, identifiable production batch data in the blockchain network includes: Extract key process parameters from the production batch data that has been put on the chain, including solder paste printing thickness, chip placement accuracy, and reflow soldering temperature profile parameters; Calculate the statistical characteristic values ​​of key process parameters, including mean, variance, extreme values, and trend slope; Quality inspection indicators are extracted from production batch data, including solder joint quality rating, component missing rate, and defective component rate; Statistical feature values ​​and quality inspection indicators are combined into a multidimensional feature vector, and the feature vector is then dimensionality-reduced to retain the most discriminative feature dimensions.

6. The blockchain-based SMT production data traceability method according to claim 5, characterized in that, The step of binding the data feature vector with the corresponding production batch identifier and storing it in the Merkle tree data structure of the blockchain includes: Establish a mapping table between data feature vectors and production batch identifiers, and store the mapping table in the blockchain after hashing. The data feature vector itself is compressed and encrypted, then stored in a distributed file system, and the corresponding file hash address is saved in the blockchain; The logic of associating data feature vectors with production batch identifiers is implemented through smart contracts to ensure the integrity and security of data queries. Regularly perform integrity checks on the stored data feature vectors, and verify whether the data has been tampered with by comparing the hash values ​​stored in the blockchain.

7. The blockchain-based SMT production data traceability method according to claim 6, characterized in that, The step of retrieving the data feature vector corresponding to a specific production batch identifier from the blockchain includes: Based on the query conditions, the range and time interval of the production batch identifier to be retrieved are analyzed; Locate the block and transaction records containing the target production batch identifier through the blockchain's indexing service; Retrieve the corresponding data feature vector ciphertext from the distributed file system, and decrypt and decompress it using the authorization key; The decrypted data feature vectors are parsed to restore them to an analyzable structured data format.

8. The blockchain-based SMT production data traceability method according to claim 7, characterized in that, The pattern recognition analysis of the data feature vectors includes: The data feature vectors are input into a trained production process evaluation model, which is trained based on historical normal production data. Calculate the similarity measure between the input data feature vector and the normal production mode feature vector; Based on the magnitude of the similarity metric, the degree of deviation between the current production process and the normal pattern is determined; Multiple threshold ranges are set to classify the health status of the production process into multiple levels based on the degree of deviation.

9. The blockchain-based SMT production data traceability method according to claim 8, characterized in that, The process for initiating traceability of relevant production batch data includes: Based on the anomaly level obtained from pattern recognition analysis, the range of historical production batches that need to be traced is determined; Based on the pattern characteristics of the abnormal features, identify the key process steps and quality parameters that require special attention; Tracing the data feature vectors of relevant production batches in reverse chronological order along the production data records stored in the blockchain; By comparing the differences in data feature vectors between abnormal production batches and normal production batches, key process parameters that may lead to the abnormality can be identified.

10. The blockchain-based SMT production data traceability method according to claim 9, characterized in that, After determining the depth and scope of the tracing, the process also includes: Based on the identified key process parameters, query the blockchain for other production batch records that use the same process parameters; Analyze the frequency and distribution patterns of abnormal process parameters to identify systemic process problems; The results of the traceability analysis are stored in the blockchain in the form of a structured report, including an assessment of the root cause of the anomaly and the scope of its impact; Based on the traceability analysis results, the process parameter settings of relevant production equipment are automatically adjusted, and the parameter adjustment events are recorded in the blockchain.

Images