A method for reconstructing and predicting key logging sequences of a sanding well of an oil and gas well

Abstract

The application discloses a kind of oil and gas well sanding well key well logging sequence reconstruction and prediction method, it is related to oil and gas field development technical field, comprising the following steps: S1, selected target block contains the modeling well of complete well logging sequence curve, and based on pears correlation, the correlation of well logging key sequence curve and well logging other series curve in complete well logging sequence curve is quantitatively evaluated;S2, based on radial basis neural network model and correlation quantification evaluation result, establish the nonlinear relationship prediction model of sanding key parameter well logging sequence and train prediction model;S3, using prediction model, the well logging sequence curve reconstruction of the target well of block is carried out;S4, using the reconstruction well logging sequence curve of target well, sanding prediction is carried out to target well.The present application makes full use of the correlation between actual data and data, to establish suitable neural network prediction model and carry out sanding prediction to target well, and it has important practical significance to the evaluation of domestic sanding prediction.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, and more specifically to a method for reconstructing and predicting key logging sequences in oil and gas wells producing sand. Background Technology

[0002] In current oil and gas well production, sand production is the most common and serious problem faced in reservoir production. Sand production risk prediction is one of the key technical means to effectively prevent sand production, and the logging sequence used for sand production prediction is the most important basic research data. However, due to historical reasons and economic factors, most wells in a block do not have complete logging sequences, so it is necessary to reconstruct certain logging sequences that play a crucial role in sand production prediction.

[0003] However, all current well logging sequence predictions are based on rock physics models, which can be called "model-driven" data prediction methods. This method has some shortcomings. For example, rock physics models themselves are very complex, with many parameters, making it difficult to correlate them one-to-one with actual formation data, thus leading to inaccurate well logging prediction results; rock physics models are based on laboratory test data, resulting in inaccurate data points; and the labor costs are high, requiring several days to complete the well logging prediction for a single well.

[0004] Therefore, new technologies and methods are needed to improve the accuracy and economic benefits of well logging sequence prediction. Summary of the Invention

[0005] To overcome the shortcomings of the existing technology, this invention discloses a method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells. Based on key parameters for sand production prediction, this invention makes full use of the correlation between actual data and data to establish a suitable neural network model for predicting sand production in wells lacking key information for sand production prediction.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for reconstructing and predicting key logging sequences in oil and gas wells producing sand includes the following steps:

[0008] S1. Select modeling wells in the target block that contain complete logging sequence curves, and quantitatively evaluate the correlation between the key logging sequence curves and other logging series curves in the complete logging sequence curves based on Pearson correlation.

[0009] S2. Based on the radial basis function neural network model and the correlation quantification evaluation results, a nonlinear relationship prediction model of the logging sequence of key sand production parameters is established and the prediction model is trained.

[0010] S3. Using the prediction model, reconstruct the logging sequence curves of the target wells in the block;

[0011] S4. Use the reconstructed logging sequence curves of the target well to predict sand production in the target well.

[0012] I. Quantitative Evaluation of the Correlation Between Key Sequences and Other Well Logging Series

[0013] S1. Select modeling wells in the target block that contain complete logging sequence curves, and quantitatively evaluate the correlation between the key logging sequence curves and other logging series curves in the complete logging sequence curves based on Pearson correlation.

[0014] Preferably, in step S1, the complete logging sequence curve includes neutron, resistivity, well diameter, gamma, potential, density, acoustic wave, and shale content.

[0015] The key logging sequence curves include density, acoustic wave, and shale; the other logging sequence curves include neutron, resistivity, borehole diameter, gamma, and potential.

[0016] Preferably, in step S1, the complete logging sequence curve is analyzed and data quality is controlled, including checking for outliers and unreasonable data.

[0017] Preferably, in step S1, the Pearson correlation coefficient of the Pearson correlation is:

[0018]

[0019] Where r(X,Y) is the Pearson correlation coefficient; X and Y represent two samples, namely the key sequence curve of well logging and other series curves of well logging, X={x1,x2,...,x3} n},Y={y1,y1,..,y n}; These are the means of X and Y, respectively; x i y i These are values ​​from X and Y, respectively; n is the number of samples.

[0020] Preferably, in the Pearson correlation of step S1:

[0021] Take the absolute value of the Pearson correlation coefficient, which ranges from 0 to 1. The larger the absolute value, the stronger the correlation.

[0022] When 0 ≤ absolute value < 0.2, it indicates a very weak correlation or no correlation.

[0023] When 0.2 ≤ absolute value < 0.4, it indicates a weak correlation;

[0024] When 0.4 ≤ absolute value < 0.6, it indicates a moderate degree of correlation;

[0025] When 0.6 ≤ absolute value < 0.8, it indicates a strong correlation;

[0026] When 0.8 ≤ absolute value ≤ 1, it indicates a very strong correlation.

[0027] II. Establishment of Prediction Model

[0028] S2. Based on the radial basis function neural network model and the correlation quantification evaluation results, a nonlinear relationship prediction model of the logging sequence of key sand production parameters is established and the prediction model is trained.

[0029] Preferably, in step S2, based on the correlation quantification evaluation results of Pearson correlation, well logging sequence curves with strong correlation are selected, and a nonlinear relationship prediction model for key sequence curves of sand logging is established through a radial basis neural network model.

[0030] Preferably, in step S2, the establishment and training of the nonlinear relationship prediction model includes the following steps:

[0031] S21. Establish a prediction model based on the radial basis function neural network model;

[0032] S22, Input training parameters;

[0033] S23. Set up the prediction model architecture;

[0034] S24. Train the prediction model error using the training parameters and analyze the error; if the error analysis meets the conditions, determine the prediction model architecture and proceed to the next step; if the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture.

[0035] S25. Input the parameter to be measured;

[0036] S26. Calculate the prediction model error using the parameters to be measured and analyze the error; if the error analysis meets the conditions, output the prediction result; if the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture.

[0037] III. Well Logging Sequence Curve Reconstruction

[0038] S3. Using the prediction model, reconstruct the logging sequence curves of the target wells in the block;

[0039] IV. Sand Production Prediction

[0040] S4. Use the reconstructed logging sequence curves of the target well to predict sand production in the target well.

[0041] Preferably, step S4 includes: calculating the rock mechanical parameters and geostress of the target well based on the reconstructed predicted logging sequence curves; calculating the load capacity on the wellbore rock according to the Mohr-Coulomb shear theory; and judging the wellbore stability risk using the wellbore stability coefficient; and predicting sand production for target wells with wellbore instability risk based on the elastic combined modulus method.

[0042] Preferably, in step S4, the load-bearing capacity of the wellbore rock is calculated based on the Mohr-Coulomb shear theory, and the wellbore stability risk is assessed using the wellbore stability coefficient, including:

[0043] The actual load borne by the rock is:

[0044] The allowable load on the rock is:

[0045] The wellbore stability coefficient is: K = [σ1] / [σ c ];

[0046] Where [σ1] represents the actual load borne by the rock; σ Max σ is the maximum principal stress; Min The minimum principal stress; P p α represents the formation pore pressure; α is the Biot elastic coefficient. The internal friction angle; [σ c ] represents the rock's allowable load-bearing capacity; c represents the cohesion; and K represents the wellbore stability coefficient.

[0047] Preferably, in step S4, when the wellbore stability coefficient K>1, the wellbore becomes unstable; when the wellbore stability coefficient K=1, the rock is in a limiting equilibrium state; and when the wellbore stability coefficient K<1, the wellbore is stable.

[0048] Preferably, in step S4, the elastic composite modulus is:

[0049]

[0050] Among them, E c ρ is the elastic composite modulus; ρ is the rock density; Δt c This refers to the time difference of sound waves.

[0051] Preferably, in step S4, the sand production prediction based on the elastic combined modulus method includes:

[0052] When the elastic combination modulus E c Satisfy E c ≥2.0×10 4 At MPa, the oil well does not produce sand during normal production;

[0053] When the elastic combination modulus Ec Satisfy 1.5×10 4 MPa < E c <2.0×10 4 At MPa, the oil well produces slight sand during normal production;

[0054] When the elastic combination modulus E c Satisfy E c ≤1.5×10 4 At MPa, the oil well produces severe sand during normal production.

[0055] The beneficial effects of this invention are:

[0056] This invention provides a new data-driven method for predicting sand production in the field, based on key parameters for sand production prediction. It makes full use of the correlation between the key sequence curve of well logging and other series of well logging curves, that is, it makes full use of the correlation between actual data and data, so as to establish a suitable neural network prediction model to predict sand production of the target well. This has important practical significance for the evaluation of sand production prediction in China. Attached Figure Description

[0057] Figure 1 This is a flowchart of the key logging sequence reconstruction and prediction method for sand-producing wells in oil and gas wells according to the present invention;

[0058] Figure 2 This is a flowchart of the prediction process of the RBF neural network model of the present invention;

[0059] Figure 3 This is a functional diagram of the logging sequence of the present invention;

[0060] Figure 4 This is a Pearson correlation coefficient diagram of the acoustic time difference in this invention;

[0061] Figure 5 This is a Pearson correlation coefficient diagram for mud logging according to the present invention;

[0062] Figure 6 This is a Pearson correlation coefficient diagram for density logging in this invention;

[0063] Figure 7 The training logging sequence curves of the modeled wells in this invention;

[0064] Figure 8 The consistency rate of the modeling well acoustic prediction results in this invention;

[0065] Figure 9 This refers to the logging reconstruction results of the target well (sonic prediction results of the target well) of this invention;

[0066] Figure 10 This refers to the target well logging reconstruction results (target well density prediction results) of this invention;

[0067] Figure 11 This is the logging reconstruction result of the target well in this invention (target clay content prediction result). Detailed Implementation

[0068] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention.

[0069] Example 1

[0070] A method for reconstructing and predicting key logging sequences in oil and gas wells producing sand, such as Figure 1 As shown, it includes the following steps:

[0071] S1. Select modeling wells in the target block that contain complete logging sequence curves, and quantitatively evaluate the correlation between the key logging sequence curves and other logging series curves in the complete logging sequence curves based on Pearson correlation.

[0072] S2. Based on the radial basis function neural network model and the correlation quantification evaluation results, a nonlinear relationship prediction model of the logging sequence of key sand production parameters is established and the prediction model is trained.

[0073] S3. Using the prediction model, reconstruct the logging sequence curves of the target wells in the block;

[0074] S4. Use the reconstructed logging sequence curves of the target well to predict sand production in the target well.

[0075] Example 2

[0076] This embodiment further elaborates on step S1 based on embodiment 1.

[0077] First, based on the logging curve sequence of the target block wells, a correlation analysis was conducted between the parameters and the key logging sequence for sand production.

[0078] Because well logging sequence curves have a certain sampling rate, and considering economic factors, only a portion of the well logging sequence curves are measured. Therefore, it is necessary to determine the correlation between existing parameters of the target block and key well logging sequence curves in order to construct a reasonable neural network model. The sand-producing well logging sequence curves include, for example, neutron (NPHI), resistivity (HLLD), well diameter (CAL), gamma (GR), and potential (SPHI). The key well logging sequence curves include data that play a decisive role in sand production prediction, such as density (Den), acoustic wave (AC), and shale (SH). It should be understood that the well logging sequence curves selected here can be called the well logging sequence curves of the modeling well, to distinguish them from the target wells of subsequent prediction. After determining the well logging sequence curves of the target well, further detection and analysis of the data can be performed, such as basic analysis and data quality control, such as checking for outliers or unreasonable data, to improve data quality.

[0079] Secondly, based on the Pearson correlation coefficient method, the correlation between the input parameters and the key parameters of the logging sequence curve is quantitatively evaluated.

[0080] The Pearson correlation coefficient is:

[0081]

[0082] Where r(X,Y) is the Pearson correlation coefficient; X and Y represent two samples, namely the key sequence curve of well logging and other series curves of well logging, X={x1,x2,...,x3} n},Y={y1,y1,..,y n}; These are the means of X and Y, respectively; x i y i These are values ​​from X and Y, respectively; n is the number of samples.

[0083] In Pearson correlation:

[0084] Take the absolute value of the Pearson correlation coefficient, which ranges from 0 to 1. The larger the absolute value, the stronger the correlation.

[0085] When 0 ≤ absolute value < 0.2, it indicates a very weak correlation or no correlation.

[0086] When 0.2 ≤ absolute value < 0.4, it indicates a weak correlation;

[0087] When 0.4 ≤ absolute value < 0.6, it indicates a moderate degree of correlation;

[0088] When 0.6 ≤ absolute value < 0.8, it indicates a strong correlation;

[0089] When 0.8 ≤ absolute value ≤ 1, it indicates a very strong correlation.

[0090] Example 3

[0091] This embodiment further elaborates on step S2 based on embodiment 2.

[0092] First, based on the characteristics of neural network models and well logging data, a reasonable neural network algorithm model is selected.

[0093] Different neural network algorithm models are compared to select the final model. Commonly used models include Backpropagation (BP), Radial Basis Function (RBF), Long Short-Term Memory (LSTM), Fully Connected Neural Network (FCNN), and Recurrent Neural Network (RNN). During model building, a portion of the well logging sequences is used as training data, and the remainder as validation data.

[0094] Well logging data is characterized by its large volume, diverse types, and significant variations with depth. Therefore, radial basis function (RBF) networks are preferred for predicting key well logging sequences due to their strong generalization ability, ability to approximate nonlinear functions with arbitrary precision, simple structure, and fast training speed.

[0095] Secondly, based on the RBF neural network model, the logging sequence parameters of the target block are trained, and a nonlinear relationship with the logging sequence of key parameters is established.

[0096] Based on the Pearson correlation coefficient analysis results, a prediction model for the key sand logging sequence curves was established using the RBF neural network model by selecting well logging sequence curves with strong correlation.

[0097] In this invention, the flowchart for establishing the RBF neural network model using MATLAB is as follows: Figure 2 As shown, it includes the following steps:

[0098] S21. Establish a prediction model based on the radial basis function neural network model;

[0099] S22, Input training parameters;

[0100] S23. Set up the prediction model architecture;

[0101] S24. Train the prediction model error using the training parameters and analyze the error; if the error analysis meets the conditions, determine the prediction model architecture and proceed to the next step; if the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture.

[0102] S25. Input the parameter to be measured;

[0103] S26. Calculate the prediction model error using the parameters to be measured and analyze the error; if the error analysis meets the conditions, output the prediction result; if the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture.

[0104] Example 4

[0105] This embodiment further elaborates on step S3 based on embodiment 3. Step S3 reconstructs the logging sequence of the target well and, based on the prediction model of the logging sequence curve, predicts and reconstructs the key logging sequences for sand production in the remaining wells.

[0106] Example 5

[0107] This embodiment further elaborates on step S4 based on embodiment 4.

[0108] In step S4, based on the predicted logging sequence curves, the rock mechanical parameters of the target well are calculated, and the geostress is calculated. Then, according to the Mohr-Coulomb shear theory, the load-bearing capacity of the wellbore rock is calculated, and the wellbore stability coefficient is used to determine its wellbore stability risk.

[0109] The actual load borne by the rock is:

[0110] The allowable load on the rock is:

[0111] The wellbore stability coefficient is: K = [σ1] / [σ c ];

[0112] Where [σ1] represents the actual load borne by the rock; σ Max σ is the maximum principal stress; Min The minimum principal stress; P p α represents the formation pore pressure; α is the Biot elastic coefficient. The internal friction angle; [σ c ] represents the allowable load on the rock; c represents the rock cohesion; K represents the wellbore stability coefficient.

[0113] Specifically, when the wellbore stability coefficient K>1, the wellbore becomes unstable; when the wellbore stability coefficient K=1, the rock is in a limiting equilibrium state; and when the wellbore stability coefficient K<1, the wellbore is stable.

[0114] Subsequently, for wells with the risk of wellbore instability, sand production prediction was carried out based on the elastic combined modulus method, thereby determining the sand production layer and well section.

[0115] The elastic composite modulus is:

[0116]

[0117] Among them, E c ρ is the elastic composite modulus; ρ is the rock density; Δt c This refers to the time difference of sound waves.

[0118] When the elastic combination modulus E c Satisfy E c ≥2.0×10 4 At MPa, the oil well does not produce sand during normal production;

[0119] When the elastic combination modulus E c Satisfy 1.5×10 4 MPa < E c <2.0×10 4 At MPa, the oil well produces slight sand during normal production;

[0120] When the elastic combination modulus E c Satisfy E c ≤1.5×10 4 At MPa, the oil well produces severe sand during normal production.

[0121] Example 6

[0122] The following examples will provide further details.

[0123] This study uses a specific well block as the research area, including one well with a complete logging sequence. The logging sequence includes neutron (NPHI), resistivity (HLLD), caliber (CAL), gamma (GR), potential (SPHI), density (Den), acoustic (AC), and shale (SH) measurements. Then, the functional characteristics of the logging sequence curves are analyzed. Figure 3 Correlation analysis was conducted between the actual logging sequence curves of the target block and the key logging sequence densities Den, sonic AC, and shale SH, using parameters such as neutron (NPHI), resistivity (HLLD), caliber (CAL), gamma (GR), and spontaneous potential (SPHI).

[0124] Wells with complete logging sequence curves were used as modeling wells. Based on the Pearson correlation coefficient method, the correlation between their logging sequence curves was obtained, and the analysis results are as follows: Figures 4-6 As shown, its acoustic wave (AC) is well correlated with GR, NP (neutron), deep lateral resistivity (HLLD), and shallow lateral resistivity (HLLS); the clay content (SH) is related to natural gamma (GR), porosity (NP), and well diameter (CAL); and the density (Den) is related to well diameter (CAL), natural gamma (GR), lateral resistivity (HLLD), and porosity (NP).

[0125] The modeled well is trained based on a radial basis function network (RBF), and the training results are as follows: Figure 7As shown, and compared with its actual well logging sequence curve, for example... Figure 8 As shown.

[0126] By training a model for the modeled well, logging sequence curves are reconstructed for other key logging sequences, such as... Figures 9-11 As shown in the figure. Finally, the predicted well logging sequence curves are used to predict sand production in the well.

[0127] The embodiments of the present invention have been described in detail above, but the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalents or substitutions are all included within the scope defined by the claims of the present invention.

Claims

1. A method for reconstructing and predicting key logging sequences in oil and gas wells producing sand, characterized in that, Includes the following steps: S1. Select modeling wells in the target block that contain complete logging sequence curves, and quantitatively evaluate the correlation between the key logging sequence curves and other logging series curves in the complete logging sequence curves based on Pearson correlation. S2. Based on the radial basis function neural network model and the correlation quantification evaluation results, a nonlinear relationship prediction model of the logging sequence of key sand production parameters is established and the prediction model is trained. S3. Using the prediction model, reconstruct the logging sequence curves of the target wells in the block; S4. Use the reconstructed logging sequence curves of the target well to predict sand production in the target well.

2. The method for reconstructing and predicting key logging sequences in oil and gas wells producing sand, as described in claim 1, is characterized in that... In step S1, the complete logging sequence curve includes neutron, resistivity, well diameter, gamma, potential, density, acoustic wave, and shale. The key logging sequence curves include density, acoustic wave, and shale; the other logging sequence curves include neutron, resistivity, borehole diameter, gamma, and potential.

3. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 1, characterized in that, In step S1, the complete logging sequence curves are analyzed and data quality is controlled, including checking for outliers and identifying unreasonable data.

4. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 1, characterized in that, In step S1, the Pearson correlation coefficient is: Where r(X,Y) is the Pearson correlation coefficient; X and Y represent two samples, namely the key sequence curve of well logging and other series curves of well logging, X={x1,x2,...,x3} n },Y={y1,y1,..,y n }; These are the means of X and Y, respectively; x i y i These are values ​​from X and Y, respectively; n is the number of samples.

5. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 4, characterized in that, In the Pearson correlation of step S1: Take the absolute value of the Pearson correlation coefficient, which ranges from 0 to 1. The larger the absolute value, the stronger the correlation. When 0 ≤ absolute value < 0.2, it indicates a very weak correlation or no correlation. When 0.2 ≤ absolute value < 0.4, it indicates a weak correlation; When 0.4 ≤ absolute value < 0.6, it indicates a moderate degree of correlation; When 0.6 ≤ absolute value < 0.8, it indicates a strong correlation; When 0.8 ≤ absolute value ≤ 1, it indicates a very strong correlation.

6. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 1, characterized in that, In step S2, based on the correlation quantification evaluation results of Pearson correlation, well logging sequence curves with strong correlation are selected, and a nonlinear relationship prediction model for key sequence curves of sand logging is established through a radial basis neural network model.

7. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 1, characterized in that, In step S2, the establishment and training of the nonlinear relationship prediction model includes the following steps: S21. Establish a prediction model based on the radial basis function neural network model; S22, Input training parameters; S23. Set up the prediction model architecture; S24. Train the prediction model error using the training parameters and analyze the error; if the error analysis meets the conditions, determine the prediction model architecture and proceed to the next step; if the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture. S25. Input the parameter to be measured; S26. Calculate the prediction model error using the parameters to be measured, and analyze the error; if the error analysis meets the conditions, output the prediction result. If the error analysis does not meet the conditions, return to step S23 to adjust the prediction model architecture.

8. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 1, characterized in that, Step S4 includes: based on the reconstructed predicted logging sequence curves, calculating the rock mechanical parameters and geostress of the target well, calculating the load capacity on the wellbore rock according to the Mohr-Coulomb shear theory, and using the wellbore stability coefficient to determine the wellbore stability risk; for target wells with wellbore instability risk, predicting sand production based on the elastic combined modulus method.

9. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 8, characterized in that, In step S4, the load-bearing capacity of the wellbore rock is calculated based on the Mohr-Coulomb shear theory, and the wellbore stability risk is assessed using the wellbore stability coefficient, including: The actual load borne by the rock is: The allowable load on the rock is: The wellbore stability coefficient is: K = [σ1] / [σ c ]; Where [σ1] represents the actual load borne by the rock; σ Max σ is the maximum principal stress; Min The minimum principal stress; P p α represents the formation pore pressure; α is the Biot elastic coefficient. The internal friction angle; [σ c ] represents the allowable load on the rock; c represents the rock cohesion; K represents the wellbore stability coefficient.

10. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 8, characterized in that, In step S4, when the wellbore stability coefficient K>1, the wellbore becomes unstable; when the wellbore stability coefficient K=1, the rock is in a limiting equilibrium state; and when the wellbore stability coefficient K<1, the wellbore is stable.

11. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 8, characterized in that, In step S4, the elastic combined modulus is: Among them, E c ρ is the elastic composite modulus; ρ is the rock density; Δt c This refers to the time difference of sound waves.

12. The method for reconstructing and predicting key logging sequences for sand-producing wells in oil and gas wells as described in claim 8, characterized in that, In step S4, the sand production prediction based on the elastic combined modulus method includes: When the elastic combination modulus E c Satisfy E c ≥2.0×10 4 At MPa, the oil well does not produce sand during normal production; When the elastic combination modulus E c Satisfy 1.5×10 4 MPa < E c <2.0×10 4 At MPa, the oil well produces slight sand during normal production; When the elastic combination modulus E c Satisfy E c ≤1.5×10 4 At MPa, the oil well produces severe sand during normal production.

Images