Quality traceability plays an essential role in assembling and welding offshore platform blocks. The improvement of the welding quality traceability system is conducive to improving the durability of the offshore platform and the process level of the offshore industry. Currently, quality management remains in the era of primary information, and there is a lack of effective tracking and recording of welding quality data. When welding defects are encountered, it is difficult to rapidly and accurately determine the root cause of the problem from various complexities and scattered quality data. In this paper, a composite welding quality traceability model for offshore platform block construction process is proposed, it contains the quality early-warning method based on long short-term memory and quality data backtracking query optimization algorithm. By fulfilling the training of the early-warning model and the implementation of the query optimization algorithm, the quality traceability model has the ability to assist enterprises in realizing the rapid identification and positioning of quality problems. Furthermore, the model and the quality traceability algorithm are checked by cases in actual working conditions. Verification analyses suggest that the proposed early-warning model for welding quality and the algorithm for optimizing backtracking requests are effective and can be applied to the actual construction process.
Offshore platforms are characterized by their intricate structure and extensive welding [
Traceability refers to the ability to trace the history, application, or location of the object under consideration [
As a result, in order to better solve the problem of welding quality traceability in the offshore platform manufacturing process, we first carried out the combing of the business process of block welding quality management and clarified the welding inspection sequence and content from the demand of quality traceability. Furthermore, we systematically summarize the elements and the instantiation stage of the Quality Traceability Information Units on the basis of the work orders. Then we provide a comprehensive quality traceability model that describes two aspects of welding quality inspection, respectively, “Follow-up Tracing” and “Problems Backtracking”. In the first aspect, we used time series analysis to establish a welding quality early warning model considering that the welding quality data complete statistics over time and is characterized by incomplete data and lack of stability. By comparing various modeling approaches, the Long Short-term Memory network (LSTM) has the ability to handle unstable time series with large sample data and can meet the practical engineering requirements for practicality and robustness. Therefore, LSTM is selected to model the quality follow-up tracing operation. In the second aspect, in order to be able to quickly query all production execution information and complete defect identification and location based on the work order number entered by the early warning model, based on statistical principles and data mainstream query algorithms, we established a quality data query optimization algorithm and a backtracking location model based on Rank Sum Test by comparing various nonparametric tests and intelligent optimization algorithms to accommodate fast and accurate finding of large sample data with no Continuous data under the welding defect discrimination. Finally, we carried out case validation under actual engineering scenarios to demonstrate the feasibility of the above method for quality traceability process modeling and its ability to solve the problems of lack of quality problem warning, slow traceability of welding problems, and manual judgment of defect types in the quality management stage of enterprises.
The remainder of the paper is structured as follows: We describe the essential aspects of welding quality traceability information in
A complicated offshore platform is usually composed of multiple blocks. In offshore platform manufacturing firms, a single block is typically utilized as a quality control node at the workshop level. Taking the block as the basic unit, by analyzing the business process of welding quality management, combing all quality inspection contents and inspection standards in the whole life cycle of blocks production, and refining various traceability information elements to work order layer granularity, so as to meet the practical application needs of enterprises for a single weld corresponding to individual. In this section, we elaborate on these elements.
The offshore platform construction adopts the mainstream integrated construction mode of design, procurement, and construction in the industry, specialized production-oriented by the assembly and welding of intermediate products, implements work order control management and compound type of work organization, establishes Lane separation process flow, and adopts parallel integrated manufacturing mode. As the core of the integrated construction technology, the blocks construction process is shown in
The project management department is primarily responsible for developing the welding quality Inspection and Testing Plan (ITP) at the initial stage of projects. The specific quality inspection procedure and precise regulations must be completed in collaboration with the technical department, the proprietor, and the third-party inspection, and must be used as a reference and standard for future welding quality inspections. Following the creation of the plan, the weld quality inspector will be taught and evaluated in accordance with the projects’ quality criteria. The quality inspector's inspection rating must be summarized and recorded in the Quality Control (QC) department.
The QC department needs to verify the supplier, inspect the welding materials and equipment, and complete the arrival quality inspection report after the examination at the stage of incoming inspection. The proprietor and the third-party inspection shall check key equipment upon its arrival. The QC department is in charge of contacting the proprietor and any third-party inspections that may be involved in the inspection process, as well as organizing all parties to finish the inspection process and provide any necessary certificates.
The QC department's main task in the production stage is to inspect the assembly and welding construction operations at the stage specified in ITP, primarily welding appearance inspection, dimensional error inspection, and nondestructive inspection, as well as conduct sampling inspection in the production process to ensure the quality of the construction operation. Specific to different operation types, such as supervision of pre-set steel plates and pipe fittings, welding material inspection, machined part inspection, cable inspection, and so on. In the manufacturing stage, the QC department's major responsibility is to examine the assembly and welding processes. Self-inspection by the manufacturing unit, mutual inspection by supervisors, and special inspection by the quality department are all part of the inspection process at each level. The proprietor and third-party inspection will perform sampling inspections as part of the quality inspection process, and the QC department will need to submit the inspection application at the project stage at the key nodes. The third-party inspectors are aided by the QC department in completing the inspection. It is critical to categorize and coordinate all departments to settle disagreements.
The QC department predominantly completes general assembly inspection, completion commissioning, launching inspection, and other tasks throughout the delivery stage. Furthermore, it is in charge of inspecting a huge number of outfitting parts. The NDT of segmented welding is mostly handled by the general assembly. Following that, the QC department is required to compile and record essential quality reports and other written records.
In the quality assurance stage, provide structural dimension displacement detection, strength detection, flaw detection. If there are any quality problems, it will track out the individual who is accountable for welds, and provide rework services and technical support for the offshore platform during the daily operation of deep-sea oil wells.
According to the analysis of the aforementioned research, a substantial amount of production data is created throughout the block construction and quality management processes for further traceability. We propose the concept of “Quality Traceability Information Unit” (QTU) based on work order to fully express the data content of the manufacturing process of the block in each link of the production line and to structurally express the data. QTUs are divided according to work orders. A single QTU includes intermediate product information, production execution information, resource information and quality inspection information related to blocks processing sets. Each QTU is associated with a work order and encompasses the three types of elements listed below.
The carrier of arranging production activities is production execution information. Due to the discontinuous manufacture of each component product in a discrete manufacturing process, the whole manufacturing process is generally split by a manufacturing bill of materials (MBOM) [
No. | Field name | Type | Content |
---|---|---|---|
1 | woNo | NVARCHAR2 | The unique work order number output according to the production plan |
2 | jobType | NVARCHAR2 | Describes the operation type, which belongs to productive work order or nonproductive work order |
3 | jobContent | NVARCHAR2 | Describe the job content of the work order |
4 | stationNo | NVARCHAR2 | Describe the production station number |
5 | processNo | NVARCHAR2 | Describe the production process number |
6 | planstartTime | DATE | Describe scheduled start time |
7 | planendTime | DATE | Describe scheduled end time |
8 | dispatchDate | DATE | Describe dispatch date |
9 | actualstartTime | DATE | Describe actual start time |
10 | actualendTime | DATE | Describe actual end time |
11 | actualworkingHour | NUMBER | Describe actual work hours |
12 | mbomNo | NUMBER | Describe MBOM number |
13 | mbomID | NUMBER | Describes the serial number in the MBOM attribute |
14 | mbomSupplier | NVARCHAR2 | Describe supplier |
15 | productionDate | DATE | Describe production date |
16 | materialCode | NVARCHAR2 | Description material code |
17 | craftCode | NVARCHAR2 | Describe craft code |
18 | qualityStandard | NVARCHAR2 | Describe quality parameter requirements |
The assembly and welding process consume a significant portion of an offshore platform manufacturing firm's production resources, including process, station, staff, material resource characteristics, etc. In order to provide optimal coordinate manufacturing resources, define the total quantity and source of all resources used from raw material input to final platform delivery, and collaborate with material attribute analysis in the quality traceability process. According to its function, manufacturing resource information is separated into process resource information, human resource information, station resource information, and material resource information. The manufacturing resource information includes the kind of process operation, welder, and process equipment and material used during blocks assembly and welding. The elements of manufacturing resource information are presented in
No. | Field name | Type | Content |
---|---|---|---|
1 | woNo | NVARCHAR2 | Unified work order number is used to associate production master data |
2 | drawingNo | NVARCHAR2 | Describe drawing number describing design input |
3 | drawingVersion | NVARCHAR2 | Describe the drawing version of the design input |
4 | craftCode | NVARCHAR2 | Describe craft code |
5 | workinstructionNo | NVARCHAR2 | Description work instruction number |
6 | workinstructionVersion | NVARCHAR2 | Describe work instruction version |
7 | changeNotification | NVARCHAR2 | Describe change notice |
8 | deviceID | NUMBER | Describe the equipment number |
9 | frockNo | NUMBER | Describe frock number |
10 | toolNo | NUMBER | Describe tool number |
11 | stationNo | NVARCHAR2 | Describe the production station number |
12 | employeeID | NUMBER | Describe employee number |
13 | employeeName | NVARCHAR2 | Describe employee name |
14 | workType | NVARCHAR2 | Describe type of work |
15 | departmentID | NUMBER | Describe the employee's department |
16 | qualification | NVARCHAR2 | Describe personal qualifications |
17 | materialCode | NVARCHAR2 | Describe material code |
18 | materialAttributes | NVARCHAR2 | Describe material properties |
19 | materialSupplier | NVARCHAR2 | Describe supplier information |
20 | materialWarranty | NVARCHAR2 | Describe material quality assurance information |
In the production and manufacturing process, as well as subsequent operation and maintenance, quality inspection feedback information is a description and record of numerous process requirements, process status, and inspection results. The structural inspection data for the whole life cycle of the block, from the start of the manufacturing line through subsequent operation and maintenance, must be collected. It primarily contains information on block structure, quality characteristics, production abnormalities, engineering changes, and quality event feedback, among other things. The quality feedback information may be traced back to the platform's quality characteristics and the relevant iterative update records in the production process, which is a crucial element of the block manufacturing process's backtracking. The quality feedback information elements are listed in
No. | Field name | Type | Content |
---|---|---|---|
1 | woNo | NVARCHAR2 | Unified work order number is used to associate production master data |
2 | itpNo | NUMBER | Describe ITP number |
3 | inspectionTime | DATE | Describe inspection time |
4 | employeeID | NUMBER | Describe the person in charge of welding |
5 | qualification | NVARCHAR2 | Describe personal qualifications |
6 | selfinspectionResult | NVARCHAR2 | Describe self inspection results |
7 | appearanceinspectionResult | NVARCHAR2 | Describe the appearance inspection results |
8 | dimensioninspectionResult | NVARCHAR2 | Describe dimensional inspection results |
9 | ndtResult | NVARCHAR2 | Describe NDT results |
10 | materialAttributes | NVARCHAR2 | Describe material properties |
11 | weldmaterialAttributes | NVARCHAR2 | Describe welding material properties |
12 | weldcraftInfo | NVARCHAR2 | Describe welding craft information |
13 | craftCode | NVARCHAR2 | Describe craft code |
14 | qualityStandard | NVARCHAR2 | Describe quality parameter requirements |
15 | warrantydisplacementResult | NVARCHAR2 | Describe the inspection results of dimension displacement |
16 | warrantystrengthResult | NVARCHAR2 | Describe the quality assurance strength test results |
17 | warrantyflawdetectionResult | NVARCHAR2 | Describe the quality assurance flaw detection results |
Each unit is associated with the stages of design, plan, progress, quality inspection, materials, outsourcing, operation, and maintenance, as can be observed by combining the above QTU elements. As a consequence, before constructing the traceability model, the functions of each information module must be clarified. In
To accomplish traceability, it is necessary to establish a traceability connection between QTU and the company's existing information systems. In this part, we create an entity-relationship model based on the information business research and traceability model requirements, which define and describe what is important to processes in quality traceability [
In this section, we propose a set of composite welding quality traceability models that may be easily adapted to the actual offshore platform blocks construction process. Quality traceability is the process of identifying and then addressing faults. As a consequence, the most frequent errors in the blocks welding process should be analyzed first. Welding defects are the most prevalent cause of structural failure and accidents [
The primary objective of the welding quality tracking process is to follow inspection results over time and determine if the welding quality is abnormal based on these data models, which can be done using multivariate time series prediction technology. In contrast to commonly used prediction models (such as Regression Analysis, Grey Prediction, and Markov Prediction), time series analysis realizes the prediction function by fitting and parameter estimation according to the measured data, eliminates the overfitting phenomenon of Regression Analysis in the face of high-order data, solves the lack of computational power of the Grey Prediction model when dealing with large sample data, and avoids the high requirements of a Markov Prediction model on the integrity and stability of enterprise quality data.
In multivariate time series analysis, anomaly identification is a critical issue. Many researchers have used deep learning algorithms to cope with anomaly detection in complicated data in recent years [
The LSTM-based welding quality early-warning model is primarily used to input the welding strength, shape variable, stress value, and other parameters of all finished structural components, as well as to compute the fluctuation of relevant parameters in real-time. For instance, when the welding strength prediction curve clearly deviates from the welding process standard, an early-warning signal is issued to inform the next stage of the backtracking inquiry. In order to fulfill high-performance time series data prediction while keeping the quality tracking process interpretable. Following the model's generation of tracking prediction data, the residual is evaluated using a non-parametric and dynamic threshold approach, and the related work order is queried through the associated database and output to the quality problem backtracking section to begin the relevant process.
To be more specific, we create a set of models based on the number of work orders and welding inspection indications. Each work order gets its own model, which is used to store detection and prediction results. In
We assume that
In
In
In order to reduce the possible false positives of quality defects in the model, based on the research of [
A quality problem backtracking signal is represented as
The Quality Problems Backtracking procedure is initiated based on the Quality Follow-up Tracking model after getting the welding defect early-warning signal or manually submitting the application. The main principle behind modeling in the enterprise quality management process is to concentrate on the breadth and speed of backtracking search, immediately clarify the search direction, and identify groupings.
The complete quality problem backtracking model contains two processes, which are quality problem backtracking query and defect identification. In the actual quality traceability process of offshore enterprises, the huge number of data tables and scattered key information, usually lead to slow query speed and inaccurate output information. Establishing a quality data query optimization algorithm is a reasonable solution to the exponential growth of data tables with the increase in query plans. Among the current mainstream query algorithms, there are three main categories, which are the Enumeration method, Heuristic Algorithms, and Intelligent Optimization Algorithms. Enumeration method as the query algorithm with the highest accuracy, by traversing all the data information, can completely output all the doubtful work order information, but its low efficiency is difficult to meet the fast query requirements. Heuristic algorithms are generally based on priority rules, which reduce the complexity of the search space by polynomials, but usually fall into the case of local optimum and cannot guarantee the stability of search results. Intelligent optimization algorithms have obvious advantages over the first two in solving query optimization problems of large databases, so genetic algorithms are chosen as the basis of query optimization algorithms in intelligent optimization algorithms, which can quickly complete the task of backtracking queries.
The defect identification process retrieves the attributes, design drawings, quality inspection sheets, production progress records, quality inspection sheets, welder information, welding materials, equipment, and other information of all welds involved under the work order according to the WO number output by the early warning model, starting with the early warning and information obtained from the query. Then, using statistical methods, connect to the enterprise's process route library, where the block welding standards are stored, extract the welding material attributes, strength attributes, weld size standards, and other data from the work order, and compare them to the work order inspection results one by one. Based on the layered method, check the defect location first, then compare the welding specifications, and determine the defect type based on the sample inspection. Finally, according to the welding execution process records, confirm that the defect belongs to one or more of the five categories: steel welding materials, mechanical equipment, welding personnel, operating environment, and process methods, and put forward the rework decision information.
The data interleaving phenomena are prevalent at this stage. Based on non-parametric tests of statistical analysis methods, the fast-processing capability and stable performance of Rank Sum Test (RST) become more important for such discontinuous data. Compared with the chi-square test, binomial distribution test, and variable randomness test, RST does not depend on the specific form of the overall distribution and can be applied without considering what kind of distribution the weld defects are and whether the distribution is known. It is very suitable for locating quality problems retrospectively. As a result, we propose an RST-based backtracking location model. Following that, it is confirmed that the defect cause is caused by one or more variables such as welding equipment and welding materials, based on the defect cause, by sample verification of five random welds under the work order. Finally, finish one by one the inventory of equipment and materials, as well as the retracing and placing of quality faults. We created a welding quality problem backtracking process model, as illustrated in
The requirements of finding defects and rapid positioning in the process of welding quality management can be realized through the integration of the quality data backtracking query algorithm and the backtracking positioning test based on RST, which forms a feasible quality problem backtracking model.
Among them, the backtracking concept is built upon a judgment approach for welding flaws. We utilize the Kruskal Wallis test (K-W test) to see if more than two samples originate from the same probability distribution, which is based on statistical principles. We'll quickly outline the process of picking five weld sampling data for rank sum inspection using the quality inspection retracing triggered by a work order as an example.
Step 1: Make a hypothesis and order all of the observed values for each sample in ascending order.
Step 2:
Step 3: Calculate the multiple Rank Sum Test statistics, and its value is:
In
Step 4: According to the saliency level and degree of freedom of the sample, the sample rejection area is obtained to determine which type of quality problem the weld defect sample belongs to.
This section describes in detail the training and algorithm implementation of the quality information early-warning model based on LSTM in the quality follow-up tracking stage, and the quality data backtracking query optimization algorithm based on Binary Search Tree (BST) in the quality problem backtracking stage in order to successfully apply the proposed set of composite welding quality traceability models in the assembly and welding process of blocks. The traceability model is linked to the production business through model training and query optimization algorithm development, and it may be enlarged and utilized in the enterprise's existing production execution system.
The execution judgment condition of the early warning model based on LSTM depends on
To train a unified model
According to the analysis of the above elements of the quality traceability unit, the quality data are scattered in each production subsystem, and the problem of data distribution and storage needs to be solved. Through model training using the federal learning paradigm, its characteristic learning objectives can be described as follows:
Among them, weights
After training all quality data source models
In
The depth neural network and federal average algorithm are utilized to begin the iteration after getting the training goal
Two LSTM layers, two Dropout layers, one Sense layer, and one Linear layer compensate the algorithm flow. The predicted time series data is generated by entering multivariate time series data. The LSTM and Dropout layers are maintained frozen during iterative computation, whereas the Sense and Linear layers are mostly modified. The following is the flow of the algorithm:
Input: Public data set
Step 1: Train the LSTM model
Step 2: Determine the number of algorithm cycles and enter the iterative calculation.
Step 3: Distribute the public data set
Step 4: Each data source trains its own model
Step 5: Aggregate and align the model parameters returned from all quality data sources, update and obtain
Step 6: End the cycle.
Step 7: Update each data source according to
Output: Quality data source model
We examine the techniques of welding defect judgment and blame localization based on rank sum test in the process of quality problem backtracking. In this section, we propose a quality information query optimization method to supplement the breakpoints from getting an early warning signal to finishing the defect diagnostic process, and we optimize the quality data traceability query path to achieve the best efficiency in the quickest period.
The strategy for executing query instructions in a database is a crucial technology for achieving data multi-join quick queries, and it is a significant component to consider when testing the database system's performance. Each input query expression will entail numerous data tables when dealing with a huge amount of welding quality data. The complexity and variety of Query Execution Plans (QEP) are determined by the diversity of connection order across tables. The algorithm must search in the subspace that may exist in the query value for extensive QEPs combined constitute a vast strategy space.
Based on the formal expression of mutating QEP into a join tree proposed by Schneider et al. [
There are I/O costs, storage costs, computing costs, and communication costs in data interactive communication. In the process of multi-connection query of welding quality data, I/O cost is the main influencing factor. We establish a model by using a dynamic cost estimation method. Taking a connection
If a query contains
In
The purpose of establishing a data multi-join query cost model is to specify the constraints for the welding quality traceability query optimization algorithm's implementation.
Through the construction of quality backtracking model and data multi-connection query cost model, the pre constraints and expected objectives of the algorithm are determined. We design an improved Genetic Algorithm based on Partition and Neighborhood-search to solve the multi-join query optimization problem of welding quality traceability data.
Based on the previous research, Partition Algorithm, which is a common quick sorting method, divides the query optimization stage into two stages according to the relational connection attributes, so as to improve the parallelism and speed up the query speed. In the first stage, it divides the sub-targets into “clusters” according to the query ability of the data source to form a new column. In the first “cluster”, all sub-targets output variables
Based on the above ideas, we design an improved flow based on genetic algorithm, as shown in
The specific process is:
Step 1: Initialize algorithm parameters.
Step 2: Encode based on the left deep tree to form a legal chromosome.
Step 3: Use the “clustering” idea of the Partition Algorithm to find an effective QEP as the initial population.
Step 4: Calculate the fitness value of the current individual and determine the termination condition, if it meets, go to Step 8, otherwise, go to Step 5.
Step 5: Perform selection, crossover, and mutation operations within the current population.
Step 6: Determine whether the calculation reaches the predetermined time interval
Step 7: Perform the neighborhood search stage.
Step 7.1: Calculate fitness value
Step 7.2: Calculate
Step 7.3: Select the chromosome
Step 7.4: Perform crossover work according to crossover probability to obtain new chromosomes.
Step 7.5: Select the best chromosome to replace
Step 8: The program ends, output the optimal QEP sequence, and find the welding quality data content according to QEP, and complete the traceability.
In the process of algorithm implementation, it is necessary to issue these additional statements of the key factors to make the algorithm completes.
Neighborhood structure is to limit the local search range to a small range. Specifically, when chromosomes
In the Multi-join Query Optimization Problem, the chromosome coding design is mostly based on the shape of the join tree, and different tree shapes correspond to different QEP. Considering that different join trees involve different connection relationships, and whether the tree shape is legal or not. We use the left deep tree as the search space, use the way of first root traversal, express the relationship (i.e., leaf node) with decimal numbers, and express the join operation (i.e., root node) with letters, so as to intuitively reflect the tree structure and facilitate the subsequent genetic operation.
According to the coding rules, the chromosomal code of query
Since this algorithm uses special methods to process population initialization and genetic operation, it will not lead to the generation of illegal chromosomes. Therefore, there is no need to introduce penalty function in individual fitness evaluation. The fitness function of this algorithm is:
The selection operation is based on individual fitness, and the main purpose is to avoid the loss of better genetic genes, so as to improve the global convergence. In this paper, roulette selection method [
Crossover probability
In this paper, the chromosomal code is generated by traversing the left deep tree first. Therefore, the premise of genetic operation is to ensure the legitimacy of chromosome, that is, to ensure the legitimacy of join tree. Based on this consideration, we adopt the crossover strategy of equal large and small trees to ensure the tree type legitimacy of left deep tree, that is, two subtrees with equal size and similar shape are randomly selected under the crossover probability to generate offspring through crossover operation. In view of the two problems that cross operation will lead to the interleaving of nodes and relationships, which makes the tree structure impossible to form, and the generation of duplicate genes after crossover, this paper takes the following measures to adjust the chromosome after crossover: exchange the leaf node representing the relationship with the root node representing the join operation. The duplicate genes were randomly changed to the remaining genes that did not appear. Let two join trees to be crossed. The chromosomes of
After crossover:
Obviously, there are duplicate gene bit
Select the offspring generated after crossing with a small probability to mutate, so as to make its gene mutation. In order to ensure that the chromosome can still be expressed as a legal join tree after mutation operation, two operations are proposed: random exchange of gene positions of any two decimal numbers in chromosome individuals; Take letters or numbers as separators and randomly divide them into equal size strings for gene bit exchange. The mutation results are shown in
In this section, we discuss an empirical validation of the proposed Quality Traceability Model and its solutions. Our major purpose for proposing this model is to cover all companies in the quality traceability process of an offshore platform manufacturing facility, as well as to finish the project application. In case verification, there are two areas of study. The first is to test the feasibility of an early warning model in the welding quality follow-up tracking stage, and the second is to examine the efficacy of the data multi-join query optimization technique in the quality problem backtracking stage.
We conducted model test experiments in the offshore platform manufacturing enterprise for more than half a year, collaborated with the enterprise's existing Manufacturing Execution System, completed the development of multiple quality related database interfaces, and obtained some welding quality data with permission in order to ensure the authenticity of the example verification results.
In particular, in order to fully train the quality early-warning model, we preset the model training parameters in
Parameter name | Value |
---|---|
Number of channels | 15 |
Feature dimension | 32 |
Number of training sets | 58632 |
Number of test sets | 80725 |
Training period | 20 |
Through the training of the Quality Early-warning Model, we have preliminarily applied it in the manufacturing enterprises of offshore engineering platforms. In this study, we have completed the collection and analysis of multiple groups of quality tracking data. Considering that there may be quality fluctuation in different areas of the block, we have selected three scene statistical models to predict the accuracy, as shown in
Date | Parts welding | |||
---|---|---|---|---|
8/16 | 79 | 1 | 12.66 | 100 |
8/17 | 165 | 0 | 0 | 0 |
8/18 | 89 | 1 | 11.24 | 50 |
8/19 | 24 | 1 | 41.67 | 100 |
8/20 | 56 | 1 | 17.86 | 100 |
8/23 | 88 | 1 | 11.36 | 33.3 |
8/24 | 60 | 2 | 33.33 | 100 |
8/25 | 34 | 0 | 0 | - |
8/26 | 101 | 0 | 0 | 0 |
8/27 | 168 | 2 | 11.9 | 66.67 |
Avg. of accuracy | 61.11% | |||
Date | Main structural members welding | |||
8/16 | 40 | 1 | 25 | 100 |
8/17 | 99 | 0 | 0 | 0 |
8/18 | - | - | - | - |
8/19 | 68 | 1 | 14.71 | 100 |
8/20 | - | - | - | - |
8/23 | - | - | - | - |
8/24 | 68 | 2 | 29.41 | 66.67 |
8/25 | 49 | 0 | 0 | - |
8/26 | 60 | 2 | 33.33 | 100 |
8/27 | - | - | - | - |
Avg. of accuracy | 73.33% | |||
Date | Blocks assembly welding | |||
8/16 | 60 | 1 | 16.67 | 50 |
8/17 | 70 | 0 | 0 | - |
8/18 | 56 | 1 | 17.86 | 100 |
8/19 | - | - | - | - |
8/20 | 50 | 0 | 0 | - |
8/23 | 87 | 1 | 11.49 | 100 |
8/24 | - | - | - | - |
8/25 | - | - | - | - |
8/26 | 25 | 0 | 0 | 0 |
8/27 | 60 | 2 | 33.33 | 100 |
Avg. of accuracy | 70.00% |
Based on the first verification results, we continuously tracked the welding quality data of parts for 3 months. The verification results are shown in
Date | Parts welding | |||
---|---|---|---|---|
8/16 | 79 | 1 | 12.66 | 100 |
8/17 | 165 | 0 | 0 | 0 |
8/18 | 89 | 1 | 11.24 | 50 |
8/19 | 24 | 1 | 41.67 | 100 |
8/20 | 56 | 1 | 17.86 | 100 |
8/23 | 88 | 1 | 11.36 | 33.3 |
8/24 | 60 | 2 | 33.33 | 100 |
8/25 | 34 | 0 | 0 | - |
8/26 | 101 | 0 | 0 | 0 |
8/27 | 168 | 2 | 11.9 | 66.67 |
Avg. of accuracy | 61.11% | |||
9/6 | 60 | 2 | 33.33 | 100 |
9/7 | 49 | 0 | 0 | 0 |
9/8 | 23 | 0 | 0 | - |
9/9 | 88 | 1 | 11.36 | 50 |
9/10 | 34 | 2 | 58.82 | 100 |
9/13 | 69 | 1 | 14.5 | 100 |
9/14 | 90 | 1 | 11.11 | 100 |
9/15 | 31 | 0 | 0 | - |
9/16 | 32 | 0 | 0 | - |
9/17 | 39 | 0 | 0 | 0 |
Avg. of accuracy | 64.29% | |||
10/11 | 39 | 0 | 0 | - |
10/12 | 45 | 0 | 0 | - |
10/13 | 56 | 1 | 17.86 | 100 |
10/14 | 44 | 0 | 0 | - |
10/15 | 40 | 1 | 25 | 100 |
10/18 | 70 | 3 | 45.86 | 75 |
10/19 | 56 | 1 | 17.86 | 100 |
10/20 | 83 | 0 | 0 | 0 |
10/21 | 115 | 2 | 17.39 | 66.67 |
10/22 | 39 | 0 | 0 | - |
Avg. of accuracy | 73.61% |
In order to verify the performance of the Data Multi-join Query Optimization Algorithm in the process of welding quality data backtracking query, we implemented the algorithm in Python and compared its performance with Partition Algorithm and Genetic Algorithm to demonstrate the feasibility and effectiveness of the algorithm in practical engineering. In the validation initialization stage, we selected 8 groups of queries, and the number of relationships increased in turn, each group ran 20 times, took the average execution time, and all tested the optimal solution and optimal convergence algebra. In the initialization part of the algorithm, the population size is set to 100, the crossover probability is 0.6, and the mutation probability is 0.01. The specific experimental results are shown in
Algorithm name | No. of relationships | Average convergence generations | Average execution time (min) | Optimal solution | Average query time per relationship (min) | Average fit of the optimized solution to the actual value |
---|---|---|---|---|---|---|
Data multi-join query optimization algorithm | 6 | 896 | 0.038 | 254.60 | 0.036 | 95.00% |
10 | 1273 | 0.062 | 577.30 | |||
15 | 1609 | 0.203 | 840.61 | |||
20 | 1962 | 0.334 | 1098.84 | |||
25 | 2338 | 1.167 | 1785.72 | |||
30 | 2721 | 1.961 | 2463.89 | |||
40 | 3524 | 2.698 | 3401.72 | |||
50 | 4342 | 3.457 | 4364.91 | |||
Partition algorithm | 6 | 855 | 0.031 | 254.60 | 0.076 | 99.12% |
10 | 2004 | 0.053 | 577.30 | |||
15 | 2900 | 0.301 | 816.30 | |||
20 | 3796 | 0.548 | 1055.30 | |||
25 | 3435 | 2.164 | 1634.01 | |||
30 | 3075 | 3.780 | 2212.72 | |||
40 | 4505 | 6.349 | 3066.55 | |||
50 | 5935 | 8.917 | 3920.38 | |||
Genetic algorithm | 6 | 889 | 0.037 | 254.60 | 0.047 | 75.17% |
10 | 2579 | 0.062 | 602.50 | |||
15 | 2719 | 0.252 | 981.36 | |||
20 | 2860 | 0.441 | 1360.22 | |||
25 | 3381 | 1.598 | 2199.08 | |||
30 | 3902 | 2.755 | 3037.94 | |||
40 | 4464 | 3.497 | 4308.64 | |||
50 | 5027 | 4.238 | 5579.33 |
It can be seen from the verification results. Each of the three algorithms can obtain its own optimal solution, and the partition algorithm can receive the optimal solution in the three algorithms. By comparing with the results of all screening by enumerating method, the partition algorithm and the algorithm proposed in this paper can better fit the actual value. According to the average convergence generations, the data multi-join query optimization algorithm converges fastest among the three algorithms, and the average query time per relationship is significantly better than the other two algorithms. Considering the principle of being fast and accurate in the actual quality traceability process, when enumeration screening cannot be implemented, we take the optimal solution of partition algorithm as the target value and compare the fitting degree of genetic algorithm and data multi-join query algorithm to the optimal solution, as shown in
In this paper, we propose a Composite Welding Quality Traceability Model to adapt to the actual construction process of offshore platform blocks, which is feasible and easy to use. By combing the business processes of offshore platform construction enterprises and summarizing the elements of the quality traceability model, the construction objectives of quality traceability model are defined, and the model is divided into two sections: Quality Follow-up Tracing and Problem Backtracking. In the stage of welding quality follow-up tracking, we put forward a set of welding Quality Early-warning Model based on LSTM, and carried out model training and case verification. It has proven that effectiveness of the proposed early-warning model in practical application. In the quality problem backtracking stage, the implementation of the Welding Quality Backtracking Query Optimization Algorithm based on BST is studied. In addition, in order to realize the source location of the quality problems, a welding defect discrimination model based on Rank Sum Test is established.
In the process of verifying the welding quality prediction and data query, the welding defect early warning with good accuracy and fast and accurate data query fully demonstrate that the above methods are suitable for establishing the welding quality management model, It can solve the low-efficiency problem found in the investigation of the actual quality management implementation process of offshore engineering enterprises, which is “result-oriented, according to the unqualified structure inspection, manual tracing of paper files, backward reasoning of the causes of welding defects and accountability”. The intelligent idea is introduced into the implementation process of quality control. Based on this series of research work, a quality traceability management system for the whole production process has been established. In comparison to the traditional quality management mode, the early warning model can assist quality inspectors in detecting welding defects earlier, the quality data backtracking query has more targeted retrieval and more efficient defect discrimination, changes the way of manual classification and search of multiple databases in the actual traceability process, and can quickly generate quality control decision information.
We will broaden the scope of existing research and address the unification of quality management and welding standards for the multi-agent dispersed maritime manufacturing sector in the future research project. We will continue to research big data and blockchain technology's vast quality data retrieval algorithms, and train a high-accuracy welding quality early warning model in line with actual enterprise production over time.
The authors are responsible for the contents of this publication. Besides, the authors would like to thank lab classmates for contribution to the writing quality.