Skip to content
BY 4.0 license Open Access Published by De Gruyter March 4, 2022

Automatic recognition method of installation errors of metallurgical machinery parts based on neural network

  • Hailong Cui EMAIL logo and Bo Zhan

Abstract

The installation error of metallurgical machinery parts is one of the common sources of errors in mechanical equipment. Because the installation error of different parts has different influences on different mechanical equipment, a simple linear formula cannot be used to identify the installation error. In the past, the manual recognition method and the touch recognition method lack of error information analysis, which leads to inaccurate recognition results. To improve the problem, an automatic recognition method based on the neural network for metallurgical machinery parts installation error is proposed. According to the principle of automatic recognition of installation error based on the neural network, the nonlinear relation matrix between layers of the neural network is established. The operating state parameters of mechanical equipment are calculated, and the time series of the parameters are divided into several segments averagely. Based on the recognition algorithm, the inspection steps of depth, perpendicularity and center position of reserved hole, base board construction, short-circuit motor line and terminal installation, center mark board, and reference point installation are designed. The experimental results show that the recall rate of the proposed method is 97.66%, and the maximum absolute deviation is 0.09. The experimental data verify the feasibility of the proposed method.

1 Introduction

The installation of metallurgical machinery refers to the process of transporting the equipment from the manufacturer to the construction site, then assembling the metallurgical equipment at the construction site, and then using it after debugging [1]. In this process, there will be many problems caused by many irresistible factors. The construction conditions of metallurgical machinery and equipment are relatively poor, so it is easy to have equipment failure in the process of equipment operation. Metallurgical machinery and equipment are different from ordinary electrical equipment. In the construction process, metallurgical machinery is a closed construction, so the failure usually occurs inside the equipment [2,3]. Generally, it is not easy to find. In addition, the structure of the equipment itself is very complex, and the cause of the fault will have certain variability, so the staff cannot quickly judge the location and the cause of the fault, which largely causes the difficulty of maintenance [4,5,6]. Therefore, to ensure a more stable and safe operation of metallurgical machinery and equipment, the relevant maintenance personnel should also have a more rigorous and in-depth understanding of the structure of the equipment, and master the maintenance technology skillfully, so that they can make accurate and rapid judgment when the fault occurs and promote the long-term development of the metallurgical industry. In the operation of metallurgical machinery and equipment, a common fault is high-temperature fault. In the detection of this fault, the staff can understand the temperature change of the relevant subsystem by touching. When the temperature is higher than the standard value, it can be determined that the subsystem has a fault [7]. This detection method can determine whether the pipeline leaks and whether the electronic devices are burned out, but the pipeline gas in the equipment may have certain toxicity, which has a negative impact on the health of workers and has a high lag. To accurately and effectively realize the prediction and the control of assembly deviation of aircraft thin-walled parts, Liu et al. propose a modeling and simulation analysis method of assembly deviation of aircraft thin-walled parts [8]. According to the assembly process of aircraft thin-walled parts, the part manufacturing deviation, tooling fixture deviation, and assembly deformation are considered. The assembly deviation model is established based on the deterministic positioning method and influences the coefficient method. On this basis, the assembly deviation model is simulated and solved based on the Monte Carlo method, and the assembly deviation is optimized according to the sensitivity of the deviation source. Zhang put forward the error analysis model method of sheet metal assembly [9]. Based on engineering experience and mathematical derivation, the multivariate first-order autoregressive model and multivariate partial linear model of sheet metal assembly process are established. The parametric and nonparametric estimates of the established model are given based on the maximum likelihood estimation method and the least square kernel smoothing estimation method. However, the aforementioned methods cannot accurately calculate the operating parameters of mechanical equipment, resulting in inaccurate information acquisition of installation error identification of different parts.

Hence, the automatic recognition method of metallurgical machinery parts installation error based on the neural network is proposed.

2 Automatic recognition principle of installation error based on the neural network

The first level network represents the candidate frame of the automatic recognition of the installation errors of the metallurgical machinery parts. The second level network represents the further enhancement of the candidate frame of the automatic recognition of the installation errors of the metallurgical machinery parts. The third level network represents the automatic recognition of the installation errors of the metallurgical machinery parts The structure was screened, and the candidate box was further regressed [10].

When inputting the first recognition image, to cope with different structural parts, it is necessary to scale the speculated image to obtain the image scale set [11,12]. The inference flow of each image is shown in Figure 1.

Figure 1 
               Automatic identification process.
Figure 1

Automatic identification process.

Figure 1 shows that the automatic recognition of installation errors of metallurgical machinery equipment parts is inferred through different levels of the network in turn, and the recognition structure of candidate parts installation errors is obtained. The nonmaximum value is consistent, and each level of the network is used as the candidate box for merging and overlapping. The size of the picture is readjusted corresponding to the candidate box of the previous level, the adjustment result is transferred to the next level for recognition again, and the recognition result is the outputted in the third level [13,14].

The main task of the second level is to filter the results of the first level output status candidates that do not conform to the actual situation. The second level network is a multitask classification network, and its structure is shown in Figure 2.

Figure 2 
               The second level network structure.
Figure 2

The second level network structure.

Figure 2 shows that, different from the first level network target, the number of candidate boxes generated in the second level is reduced, redundant data are eliminated, and regression boxes are only identified for parts installation errors. There are many neural units in the second level network structure, which greatly increases the second level network fitting [15,16,17]. Similar to the first level network, the second level network is used to identify the relative error of parts installation error. The third level is the last stage of the recognition framework, which is used to undertake the second level results and filter the redundant data. Therefore, the recognition accuracy is used as the reference result. Different from the first two levels, the third level network structure uses the full connection layer, which effectively ensures the recognition accuracy [18,19,20].

3 Wrong installation of automatic identification of metallurgical machinery parts

No matter what kind of equipment, in the process of installation, if there is an error in one step of the installation sequence, then all the following installation procedures are wrong. If it is light, it will be re-installed, and if it is serious, it will lead to the direct scrapping of the equipment and bring a lot of adverse effects to the development of the enterprise. The same is true for metallurgical machinery equipment in the process of installation, so when installing, it must be strictly followed to operate according to the correct installation steps, especially the parts of metallurgical machinery and equipment itself are extremely complex, and the equipment also has a lot of ancillary equipment, so the installation sequence must not be wrong, so as to ensure the correct installation of metallurgical equipment and so as to avoid unnecessary mistakes and bring economic losses to the enterprise. Therefore, it is necessary to combine the principle of neural network automatic recognition of installation errors to automatically identify the installation errors of metallurgical machinery parts [21].

3.1 Automatic recognition algorithm

In the automatic identification algorithm of installation errors of metallurgical mechanical equipment parts, set the following parameters. The identification time series is divided into m segments on average, n is the number of installation errors in each identification time, and k is the correlation dimension of the time series [22]. The causes and abnormal conditions are analyzed, and the equipment damage is judged according to the instruments and measuring tools. After the failure of mechanical equipment, the operation parameters change, so as to determine the fault location and cause, eliminate hidden dangers, and prevent accidents [23].

On the basis of the feature recognition of neighbor nodes, the local structure information is recognized. The role of RELU function is to increase the nonlinear relationship between the layers of the neural network. Each layer is equivalent to a matrix, which can be expressed by U:

(1) U = μ ( ω · Aggre items ( { u i z , z C ( i ) } ) + s ) .

where C(i) represents the information set of equipment parts installation, u iz vector represents the preference characteristics between mechanical equipment and parts installation, Aggreitems represents the function of mechanical equipment, μ represents the activation function, ω represents the convolution layer parameters of the neural network, and s represents the deviation.

Using undirected graph to define mechanical equipment S = (G, V, E), vertex G represents different types of data sets of mechanical equipment and parts installation, the vertex is directly connected with the weighted undirected edge V, and E represents the weight of the edge, and the formal definition is expressed as follows:

(2) E ( i , j ) = score ( i , j ) , if i A , j B , sim ( i , j ) , if i , j A or i , j B .

In formula (2), score(i, j) is the score result of mechanical equipment for parts installation and sim(i, j) is the similarity of installation process between different mechanical equipment. Therefore, the mechanical equipment is composed of the aforementioned relations, and its goal is to obtain the error identification information in the mechanical equipment.

3.2 Basic acceptance identification of equipment

Before the installation of metallurgical machinery and equipment, the foundation work should be accepted according to the designed drawings, such as the elevation and the size of the centerline of the equipment, and the embedded parts should be accepted at the same time. The depth, perpendicularity, and center position of the reserved hole should be checked. If one of these basic acceptance is unqualified, it is not in line with the acceptance standard [24]. Therefore, the corresponding rectification and treatment should be made for specific problems, and the next installation process can be carried out only after reaching the standard.

3.3 Location and identification of equipment

Equipment positioning and adjustment are mainly divided into three aspects: plane positioning, elevation adjustment, and levelness adjustment. The plane positioning of the equipment should be determined according to the previously set datum line and central standard plate. The elevation of the equipment needs to be measured, and for the more important equipment, it needs to use precision instruments for auxiliary adjustment. The adjustment of equipment levelness generally adopts the overall equipment method or split equipment method, but no matter what kind of equipment method, it needs precision instruments to make the error. The difference is minimized [25]. The process of equipment positioning and identification is shown in Figure 3.

Figure 3 
                  Equipment positioning and identification process.
Figure 3

Equipment positioning and identification process.

As shown in Figure 3, according to the process, the device positioning and identification results are output.

3.4 Construction identification of base plate

In the process of metallurgical machinery installation, in addition to checking some data of the equipment, the construction of the base plate is also an important link that needs attention in the installation process. For smooth metallurgical machinery and equipment installation, it is necessary to implement the construction of the backing plate. In the actual work, it is necessary to determine the size, the quantity, and other specific problems of the base plate according to the drawings before the installation construction and conduct in-depth research and analysis according to the layout of relevant mechanical equipment and equipment base, so as to lay a solid foundation for the later construction work. The identification process of base plate construction is shown in Figure 4.

Figure 4 
                  Identification process of base plate construction.
Figure 4

Identification process of base plate construction.

As shown in Figure 4, according to the process, the identification results of pad construction are output.

3.5 Identification of short-circuit motor circuit and terminal installation

When troubleshooting the starting motor’s weak operation, on the premise of no power problem, it has to be verified whether the wire connection is loose. If there is heat, it indicates poor contact. If the wire connection is normal, the starting motor battery is further checked. The process is shown in Figure 5.

Figure 5 
                  Motor operation inspection process.
Figure 5

Motor operation inspection process.

Tools are used to short the terminals of starter electromagnetic switch, battery, and motor conductive sheet, or to short the terminals of starter relay and battery. In addition to the short-circuit method, the battery of starter relay can be directly connected with ignition switch by wire. If the starter cannot operate normally, it indicates that there is a fault in the motor. The reason why the electromagnetic switch is faulty is that the starter cannot operate normally. Most likely, the starter and electromagnetic switch should be repaired in time. If the starter can rotate normally, it indicates that there is a fault in the circuit between the starting relay and the ignition switch, or there is a problem in the starting relay or its related circuit, which should be checked in time.

3.6 Identification of center plate and datum point

First, before the installation of metallurgical equipment, the layout plan of the location of the permanent center plate and datum point should be drawn according to the design, installation and future maintenance needs metallurgical equipment, and the number and location of the permanent center plate and datum point should be indicated in the plan, and the permanent center plate and datum point should be buried according to the location in the drawing. Punctuality is mainly used to facilitate the installation and adjustment of metallurgical equipment, which should be firmly buried and regularly inspected. Second, to observe the settlement of the whole equipment installation foundation, the foundation settlement value should be measured, solutions should be proposed on time, and settlement observation points should be set near the whole unit and main equipment. When embedding the settlement observation belt, at least four benchmark observation points should be embedded in the surrounding area of the same equipment foundation, so as to observe the settlement of the same foundation. Finally, when the equipment is installed under different conditions, auxiliary center plate and reference point can be embedded according to different needs, but the permanent center plate and reference point should always prevail.

4 Experiment

To verify the feasibility of the automatic recognition method of metallurgical machinery parts installation error based on the neural network, an experimental study was carried out.

The recognition data are set as 1,024 bits and the recall rate of data recognition is compared and analyzed by using manual recognition, touch recognition, and neural network recognition methods. The results are presented in Table 1.

Table 1

Recall rate of data identification by different methods

Parameter Manual identification Touch recognition Recognition method based on neural network
Collect all data 800 bits 840 bits 1,024 bits
Zombie data 380 bits 520 bits 24 bits
Recall ratio 52.5% 38.11% 97.66%

Table 1 shows that among the three methods, the recall rate of touch recognition method is the lowest, which is 38.11%, followed by the manual recognition method, which is 52.5%, while the recall rate of the neural network recognition method is 97.66%. According to the aforementioned comparison results, the recall rate of data recognition based on the neural network is the highest.

To further verify that the recall rate of data recognition based on the neural network recognition method is higher, it is necessary to use these three methods to collect five metallurgical machinery parts and analyze their installation error degree, as shown in Table 2.

Table 2

Comparative analysis of installation error degree of three methods

Metallurgical machinery and equipment parts Manual identification (%) Touch recognition (%) Neural network recognition method (%) Actual recognition results (%)
1 42 55 65 65
2 58 60 62 62
3 55 60 70 70
4 49 61 71 71
5 51 65 70 70

Table 2 shows that the manual recognition method and the touch recognition method are not consistent with the actual recognition results, and the neural network recognition method is consistent with the actual recognition results, which indicates that the analysis result of installation error degree using this method is more accurate.

The absolute value deviation of recognition results determines the recommendation effect. The manual recognition method, the touch recognition method, and the neural network recognition method are used to compare the absolute value deviation of recognition results. The comparison results are shown in Figure 6.

Figure 6 
               Absolute value deviation of recognition results by different methods.
Figure 6

Absolute value deviation of recognition results by different methods.

Figure 6 shows that the maximum absolute deviation of the recognition result using the manual recognition method is 0.58. The maximum absolute deviation of the recognition result using the touch recognition method is 0.61. The maximum absolute deviation of the recognition result using the neural network recognition method is 0.09. Through the aforementioned comparison results, it can be seen that the absolute value deviation of the recognition results using the neural network recognition method is low.

5 Discussion

According to the aforementioned experimental results, the proposed method has a high recovery rate of the part-related data and avoids the waste of the data information. The main reason for obtaining the ideal application performance is that the operating state parameters of mechanical equipment are calculated, and the time series of the parameters are averagely divided into several segments. Based on the recognition algorithm, the inspection steps of depth, perpendicularity, center position, base board construction, short circuit motor line and terminal installation, center mark board, and reference point installation are designed to enhance the utilization ratio of data.

Metallurgical machinery and equipment generally maintain a long-term working state, so it is prone to failure. The harm of fault is very great; therefore, it is necessary to do a good job of maintenance and provide a reasonable maintenance plan. The details are as follows:

  1. Reasonable weekly and monthly inspection plan to reduce the failure rate as much as possible.

  2. Mechanical equipment in use for a certain period of time, and it is necessary to maintain its parts, such as cleaning, painting lubricating oil, and so on. If parts are worn out and cannot be used, they must be replaced immediately.

  3. Reasonable introduction and use of automation technology to promote the improvement of equipment maintenance level.

There are many kinds of faults in the operation of metallurgical machinery and equipment. As for maintenance personnel, they must pay attention to the accumulation of daily experience and knowledge, do a good job in daily maintenance, and improve their own maintenance level. In addition, the universal meter should be used reasonably in the troubleshooting, and the troubleshooting should be completed by measuring the voltage and current of the line, so as to repair the fault in time, ensure the safety of mechanical equipment, and extend the service time.

6 Conclusion

With the development of metallurgical industry, more and more attention has been paid to the vibration, efficiency, and life of mechanical parts, and the requirement for the precision of parts installation has become higher and higher. Based on the complexity of equipment structure, we must play its rigorous working attitude in various installation links and fault diagnosis and must strictly follow the steps and processes of installation operation. In the process of operation, if a fault occurs one day, the relevant personnel should timely make an accurate diagnosis of the fault, any problems, and details to ensure the stable operation of the equipment and then to promote the long-term development of the enterprise. In this context, a neural network-based method for automatic recognition of installation errors of metallurgical machinery parts is proposed. The nonlinear relation matrix between each layer of the neural network is established. The operating state parameters of mechanical equipment are calculated, and the accurate information of installation error is obtained. Experimental results show that the proposed method has ideal data recovery rate, recall rate, and accuracy of part assembly error identification. Experimental results show that the proposed method can provide a reliable basis for related fields.

  1. Conflict of interest: The authors state no conflict of interest.

References

[1] Cheng K, Tang Q, Huang Y, Liu X, Karrow NA, Wang C. Research on recognition of medical image detection based on neural network. IEEE Access. 2020;99:1–1.Search in Google Scholar

[2] Yong L. Research on path recognition of welding manipulator based on automatic control algorithm. Acad J Manuf Eng. 2018;16(2):120–6.Search in Google Scholar

[3] Qin X, Wang P, Liu Y, Guo L, Sheng G, Jiang X. Research on distribution network fault recognition method based on time-frequency characteristics of fault waveforms. IEEE Access. 2017;6:7291–300.10.1109/ACCESS.2017.2728015Search in Google Scholar

[4] Zhu X, Zhu M, Ren H. Method of plant leaf recognition based on improved deep convolutional neural network. Cognit Syst Res. 2018;52:223–33.10.1016/j.cogsys.2018.06.008Search in Google Scholar

[5] Liang Y, Ren C, Wang H, Huang Y, Zheng Z. Research on soil moisture inversion method based on GA-BP neural network model. Int J Remote Sens. 2019;40(5–6):2087–103.10.1080/01431161.2018.1484961Search in Google Scholar

[6] Sun J, Xiao Z, Xie Y. Automatic multi-fault recognition in TFDS based on convolutional neural network. Neurocomputing. 2017;222:127–36.10.1016/j.neucom.2016.10.018Search in Google Scholar

[7] Chen Z, Cheng X, Wang X, Han M. Recognition method of dairy cow feeding behavior based on convolutional neural network. J Phys: Conf Ser. 2020;1693(1):012166.10.1088/1742-6596/1693/1/012166Search in Google Scholar

[8] Liu YJ, Tian XT, Geng JH, Zhang M, Ma LP. Modeling and simulation analysis of aircraft thin-walled parts assembly deviation. Mach Tool Hydraulics. 2020;48(2):138–43.Search in Google Scholar

[9] Zhang L. Data-driven modeling-based variation analysis for the compliant sheet metal assemblies. J Mech Eng. 2019;55(10):34–41.10.3901/JME.2019.10.034Search in Google Scholar

[10] Wang Y. Research on music recognition algorithm based on RBF neural network. Rev de la Facul de Ingenieria. 2017;32(8):707–12.Search in Google Scholar

[11] Xu H, Yan R. Research on sports action recognition system based on cluster regression and improved ISA deep network. J Intell Fuzzy Syst. 2020;39(4):5871–81.10.3233/JIFS-189062Search in Google Scholar

[12] Chen J, Zhou H, Hu H, Song Y, Gifu D, Li Y, et al. Research on agricultural monitoring system based on convolutional neural network. Future Gener Computer Syst. 2018;88:271–8.10.1016/j.future.2018.05.045Search in Google Scholar

[13] Cao C, Wu J, Zeng X, Feng Z, Wang T, Yan X, et al. Research on airplane and ship detection of aerial remote sensing images based on convolutional neural network. Sensors. 2020;20(17):4696.10.3390/s20174696Search in Google Scholar PubMed PubMed Central

[14] Hübner HB, Duarte M, Silva R. Automatic grinding burn recognition based on time-frequency analysis and convolutional neural networks. Int J Adv Manuf Technol. 2020;110(7–8):1–17.10.1007/s00170-020-05902-wSearch in Google Scholar

[15] Panda A, Sahoo AK, Panigrahi I, Rout AK. Prediction models for on-line cutting tool and machined surface condition monitoring during hard turning considering vibration signal. Mech Ind. 2020;21(5):520.10.1051/meca/2020067Search in Google Scholar

[16] Zhang X, Chen G. An automatic insect recognition algorithm in complex background based on convolution neural network. Traitement du Signal. 2020;37(5):793–8.10.18280/ts.370511Search in Google Scholar

[17] Wan K, Song JM, Kang RP. Multimodal biometric recognition based on convolutional neural network by the fusion of finger-vein and finger shape using near-infrared (NIR) camera sensor. Sensors. 2018;18(7):1315.Search in Google Scholar

[18] Jiang Q, Shen Y, Li H, Xu F. New fault recognition method for rotary machinery based on information entropy and a probabilistic neural network. Sensors. 2018;18(2):337.10.3390/s18020337Search in Google Scholar PubMed PubMed Central

[19] Yan X, Xin T. Convergence of batch gradient method based on the entropy error function for feedforward neural networks. Neural Process Lett. 2020;52(3):1–9.Search in Google Scholar

[20] Yang L, Liu Y, Peng J. An automatic detection and identification method of welded joints based on deep neural network. IEEE Access. 2019;99:1.10.1109/ACCESS.2019.2953313Search in Google Scholar

[21] Lee HG, Park G, Kim H. Effective integration of morphological analysis and named entity recognition based on a recurrent neural network. Pattern Recognit Lett. 2018;112:361–5.10.1016/j.patrec.2018.08.015Search in Google Scholar

[22] Long M, Ouyang C, Liu H, Fu Q. Image recognition of Camellia oleifera diseases based on convolutional neural network & transfer learning. Nongye Gongcheng Xuebao/Trans Chin Soc Agric Eng. 2018;34(18):194–201.Search in Google Scholar

[23] Toubhans B, Viprey F, Fromentin G, Karaouni H, Dorlin T. Study of phenomena responsible for part distortions when turning thin Inconel 718 workpieces. J Manuf Process. 2021;61:46–55.10.1016/j.jmapro.2020.11.007Search in Google Scholar

[24] Wu XS, Chen JH. Text classification on large scale Chinese news corpus using character-level convolutional neural network. J Phys: Conf Ser. 2020;1693(1):012171.10.1088/1742-6596/1693/1/012171Search in Google Scholar

[25] Wu B, Feng S, Sun G, Xu L, Ai C. Fine-grained fault recognition method for shaft orbit of rotary machine based on convolutional neural network. J Vibroengineering. 2019;21(8):2106–20.10.21595/jve.2019.20359Search in Google Scholar

Received: 2021-06-22
Revised: 2021-10-25
Accepted: 2021-12-20
Published Online: 2022-03-04

© 2022 Hailong Cui and Bo Zhan, published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 29.3.2024 from https://www.degruyter.com/document/doi/10.1515/jisys-2022-0021/html
Scroll to top button