A method for automatic identification of scale alignment in static weighing calibration of glass float
Through deep learning technology and computer control, the fully automatic calibration of the glass float meter is realized, solving the problems of manual operation risks and low efficiency in the existing technology, improving calibration efficiency and reducing errors, and achieving unattended automatic calibration process.
Patent Information
- Application Number
- CN202210763083.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-29
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-06-29
AI Technical Summary
The existing glass float calibration methods have problems such as manual operation risks, low efficiency, large errors and pollution. Especially in the static weighing method, it is necessary to manually intervene to adjust the scale and liquid alignment, which has a lot of room for improvement in efficiency.
Deep convolutional network (VGG) is used to combine industrial cameras and computer control to build a training verification data set through supervised learning and data annotation, so as to realize automatic identification of glass float meter main tube scale and calibration liquid-liquid alignment, and automatically realize the calibration process.
The full-process unmanned automatic calibration of the glass float meter is realized, saving labor costs, improving calibration efficiency, and the automatic identification accuracy reaches 98.2% and 97.4%.
Smart Images

Figure CN115144301B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of glass float calibration, in particular to an automatic identification method for scale alignment in static weighing calibration of a glass float. Background Art
[0002] A glass float is an instrument that can float vertically and freely in a liquid. The depth of the float in the liquid is used to directly measure the density, relative density or concentration of the liquid. Based on Archimedes' principle, when the glass float reaches equilibrium in the liquid, the weight of the liquid displaced is equal to its own weight. The depth of the float in the liquid can be used to directly obtain the density, relative density or concentration of the liquid from the scale.
[0003] At present, there are two main methods for calibrating glass floats: direct comparison method and static weighing method.
[0004] The direct comparison method calibrates the standard / reference density float and the float to be calibrated by immersing them in the same calibration standard solution at the same time and comparing the scale indications between them. The main drawback of this method is that laboratory staff need to prepare some toxic and harmful calibration standard solutions such as ethyl hydrogen sulfate by hand, which poses a huge risk of health hazards to the human body. At the same time, the waste liquid after the calibration will also pollute the environment. In addition, the direct comparison method requires the float to be calibrated to be read manually by the naked eye according to the lower edge of the liquid meniscus. If there are too many calibrated density points, large errors will be introduced, and it is time-consuming and labor-intensive, and the efficiency is quite low.
[0005] Based on Archimedes' law of buoyancy, the static weighing method uses a high-precision balance (or high-precision mass comparator) to measure the buoyancy of a standard solid of known mass and volume immersed in a calibration liquid, and obtains the density of the calibration liquid, which is then transferred to the density of the main tube scale of the float to be calibrated. The advantage of this method is that the full range calibration of the float can be achieved using only one calibration liquid, and the calibration efficiency is higher than that of the direct comparison method. However, its disadvantage is that although high-definition camera technology is introduced to assist in improving the alignment accuracy of the calibration scale and the liquid surface when aligning the scale of the float to be calibrated with the calibration liquid, this process usually requires manual intervention and repeated adjustments and comparisons, and the calibration efficiency still has a lot of room for improvement.
[0006] Chinese patent application number 201010539387.X discloses a petroleum density meter calibration device and method, which briefly mentions designing a reference liquid level adjustment device to automatically achieve the up and down movement of the reference liquid level by computer control, but does not emphasize how to achieve it.
[0007] A Chinese patent application with application number 201410551344.1 discloses a glass float detection and calibration device, which mentions the use of a glass float scale line recognition device that adopts image vision detection technology to automatically identify and locate the scale on the surface of the glass float. However, the patent focuses on the proposal of a calibration device and does not emphasize the method of automatic alignment of the scale. Summary of the invention
[0008] The object of the present invention is to provide a method for automatically identifying scale alignment in static weighing calibration of a glass float, which helps to achieve full-automatic calibration of the glass float and improve efficiency, in view of the above problems existing in the prior art.
[0009] The present invention comprises the following steps:
[0010] Step 1: simulate the alignment of various possible glass float gauge dry tube scales and calibration liquid levels that may occur during calibration, and record them using an industrial camera;
[0011] Step 2: Manually identify each alignment picture recorded by the industrial camera under various conditions in step 1, classify and label them according to whether the glass float dry tube scale and the calibration liquid level are aligned, and construct a training and verification data set required for supervised learning;
[0012] Step 3: Based on the training validation data set constructed in step 2, divide the training set and validation set;
[0013] Step 4: Normalize each alignment image in the training set, and normalize the validation set in the same way;
[0014] Step 5: Based on the normalized training set and validation set data, use the deep convolutional network VGG (Visual Geometry Group) to train and validate, and save the validated VGG network;
[0015] Step 6: Use the verified VGG network to automatically identify the alignment between the glass float dry tube scale and the calibration liquid surface under new conditions.
[0016] In step 1, the alignment conditions of various glass float dry tube scales and calibration liquid levels that may appear in the simulated calibration require simulating the static weighing calibration of various glass floats, including seawater density meters, petroleum density meters, sugar meters, alcohol meters, latex meters, and Baume meters, and using industrial cameras to record various simulated misaligned and aligned states of various glass float dry tube scales and calibration liquid levels.
[0017] In step 2, each alignment picture recorded by the industrial camera under various conditions in step 1 is manually identified, classified according to whether the glass float dry tube scale and the calibration liquid level are aligned, and marked into two categories: misaligned and aligned, so that one alignment picture corresponds to a unique accurate alignment result, and a training and verification data set required for supervised learning is constructed.
[0018] In step 3, the division of the training set and the validation set is to allocate the training validation data set constructed in step 2 to the training set and the validation set according to the data volume of 4:1. The training set is used for model training, and the validation set is used for evaluation after model training. In addition, the ratio of the number of images manually marked as misaligned and aligned in the training set and the validation set should be the same or similar, generally about 3:1.
[0019] In step 4, the normalization is to divide the grayscale value of each pixel at each position in each picture by 255.
[0020] In step 5, the training and verification adjust the key parameters of the model according to the performance of the model on the verification set, mainly including: the number of convolutional layers, the number of filters, the number of pooling layers, the number of fully connected layers, the training iteration cycle, and the number of samples sent to the network for training in each batch. The corresponding adjustment ranges of the key model parameters are (10, 11, 12, 13, 14, 15, 16) layers, (16, 32, 64, 128), (1, 2, 3) layers, (1, 2, 3) layers, (300, 400, 500, 600, 700, 800), and (32, 64, 128, 256), and multiple model training and evaluation are performed.
[0021] Compared with the prior art, the beneficial effects achieved by the technical solution of the present invention are: based on deep learning technology, a method for automatic identification of scale alignment in static weighing calibration of glass float is proposed, which is expected to realize full-process unattended automatic calibration of glass float in combination with computer control technology, saving labor costs and improving calibration efficiency. Experiments show that the automatic identification accuracy of the alignment of the scale of the seawater density meter main tube and the calibration liquid surface in the training set and the verification set after training can reach 98.2% and 97.4%. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 This is a schematic diagram of the misalignment between the calibration scale of the seawater density meter and the calibration liquid level;
[0023] Figure 2 This is a schematic diagram of the seawater density meter with the calibration scale to be calibrated and the calibration liquid level aligned. DETAILED DESCRIPTION
[0024] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.
[0025] Since the glass float scale to be calibrated and the calibration liquid level are not unique in the misaligned and aligned states, such as Figure 1 and 2 As shown in the figure, the seawater density meter calibration scale and the calibration liquid level are misaligned and aligned. In the figure, the seawater density meter main pipe 1, the calibration liquid level 2, and the calibration scale 3 are marked.
[0026] The specific steps of the present invention are as follows:
[0027] Step 1: Simulate the possible alignment between the seawater densitometer main tube scale and the calibration liquid level during calibration. Figure 1 and Figure 2 These are two possible situations of the alignment of the seawater densitometer dry tube scale and the calibration liquid surface recorded by an industrial camera.
[0028] Step 2: Manually identify each alignment picture recorded by the industrial camera under various conditions collected in step 1, and classify them according to whether the seawater densitometer main pipe scale and the calibration liquid level are aligned, and mark them into two categories: misaligned and aligned, so that one alignment picture corresponds to a unique and accurate alignment result, and construct a training and verification data set required for supervised learning. For this embodiment, Figure 1 It is a possible situation where the seawater densitometer main tube scale and the calibration liquid level recorded by the industrial camera are not aligned, and it is marked as misaligned; Figure 2 It is a possible situation that the seawater densitometer main tube scale and the calibration liquid level recorded by the industrial camera are aligned and marked as aligned.
[0029] Step 3. In this embodiment, the training and verification data set constructed in step 2 contains 1680 images of seawater densitometer main pipe scale and calibration liquid surface alignment, which are divided into training set and verification set according to the data volume of 4:1, that is, the training set contains 1344 alignment images, the verification set contains 336 alignment images, the training set is used for VGG model training, and the verification set is used for evaluation after VGG model training. The ratio of the number of images manually marked as misaligned and aligned in the training set and the verification set is 3:1, that is, there are 1008 misaligned images and 336 aligned images in the training set; there are 252 misaligned images and 84 aligned images in the verification set.
[0030] Step 4: Read each aligned image in the training set and normalize it by grayscale value / 255 of each pixel at each position in each image. Normalize the validation set in the same way.
[0031] Step 5, using the deep learning framework Tensorflow as the backend, using the deep learning high-level application program interface library Keras for model VGG training and verification, selecting the optimizer "Adam", the loss function "mse", and adjusting the key parameters of the model, the key parameters are determined as follows: 13 convolutional layers, 64 filters, 3 pooling layers, 3 fully connected layers, 700 training iterations, and 128 samples per batch sent to the network for training. Through training, the automatic recognition accuracy of the seawater densitometer main pipe scale and calibration liquid surface alignment in the training set and the validation set are 98.2% and 97.4% respectively.
[0032] Step 6: Use the verified VGG network to automatically identify the calibration surface of the seawater densitometer main tube and the calibration liquid under new conditions.
[0033] The above embodiments are only preferred embodiments of the present invention and cannot be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the present invention.
Claims
1. A method for automatically identifying scale alignment in static weighing calibration of a glass float, characterized in that The following steps are involved: Step 1: simulate the alignment of various possible glass float gauge dry tube scales and calibration liquid levels that may occur during calibration, and record them using an industrial camera; Step 2: Manually identify each alignment picture recorded by the industrial camera under various conditions in step 1, classify and label them according to whether the glass float dry tube scale and the calibration liquid level are aligned, and construct a training and verification data set required for supervised learning; Step 3: Based on the training validation data set constructed in step 2, divide the training set and validation set; Step 4: Normalize each alignment image in the training set, and normalize the validation set in the same way; Step 5: Based on the normalized training set and validation set data, use the deep convolutional network VGG to perform training and validation, and save the validated VGG network; Step 6: Use the verified VGG network to automatically identify the alignment between the glass float dry tube scale and the calibration liquid surface under new conditions.
2. A method for automatically identifying scale alignment in static weighing calibration of a glass float as claimed in claim 1, characterized in that In step 1, the alignment conditions of various glass float dry tube scales and calibration liquid levels that may appear in the simulated calibration require simulating the static weighing calibration of various glass floats, including seawater density meters, petroleum density meters, sugar meters, alcohol meters, latex meters, and Baume meters, and using industrial cameras to record various simulated misaligned and aligned states of various glass float dry tube scales and calibration liquid levels.
3. A method for automatically identifying scale alignment in static weighing calibration of a glass float as claimed in claim 1, characterized in that In step 2, each alignment picture recorded by the industrial camera under various conditions in step 1 is manually identified, and classified and labeled according to whether the glass float dry tube scale and the calibration liquid level are aligned, and labeled into two categories: misaligned and aligned, so that one alignment picture corresponds to a unique accurate alignment result, and a training and verification data set required for supervised learning is constructed.
4. A method for automatically identifying scale alignment in static weighing calibration of a glass float as claimed in claim 1, characterized in that In step 3, the division of training set and validation set is to allocate the training validation data set constructed in step 2 to training set and validation set according to the data volume of 4:
1. The training set is used for model training, and the validation set is used for evaluation after model training; the ratio of the number of images manually marked as misaligned and aligned in the training set and validation set is 3:
1.
5. A method for automatically identifying scale alignment in static weighing calibration of a glass float as claimed in claim 1, characterized in that In step 4, the normalization is to divide the grayscale value of each pixel at each position in each picture by 255.
6. A method for automatically identifying scale alignment in static weighing calibration of a glass float as claimed in claim 1, characterized in that In step 5, the training and verification adjust the key parameters of the model according to the performance of the model on the verification set, mainly including: the number of convolutional layers, the number of filters, the number of pooling layers, the number of fully connected layers, the training iteration cycle, the number of samples sent to the network for training in each batch, and the corresponding adjustment range of the key parameters of the model: the number of convolutional layers is 10, 11, 12, 13, 14, 15, 16 layers, the number of filters is 16, 32, 64, 128, the number of pooling layers is 1, 2, 3 layers, the number of fully connected layers is 1, 2, 3 layers, the training iteration cycle is 300, 400, 500, 600, 700, 800 times, the number of samples sent to the network for training in each batch is 32, 64, 128, 256, and multiple model training and evaluation are performed.
