CN102649159A - Online prediction system and method for density of powder injection molded blank - Google Patents

Abstract

The invention belongs to the technical field of powder injection molding, and in particular provides an online prediction system and method for powder injection molding injection blank density. The system includes an injection molding machine, an industrial CT machine, an image processing system, a sensor network system, and an artificial neural network system; the uniform powder is fed into the injection molding machine, and injected with any set of process parameters to form a green body sample ;At the moment when the injection process is completed, the sensor network automatically detects the temperature value T and pressure value P of each point inside the injection mold; the above T and P are automatically sent to the artificial neural network system, and the neural network system automatically gives the injection blank sample Predicted values of the overall grayscale H and the local grayscale L _i . The invention has the advantages of: 1. The density distribution of the injection billet is automatically predicted, which improves the efficiency. 2. The testing process does not destroy the sample, and qualified samples can still be used, saving costs. 3. The quality of injection products can be monitored in real time, so that the quality problems of injection billets can be found in time.

Description

An online prediction system and method for density of powder injection molding blank

Technical field:

The invention belongs to the technical field of powder injection molding, and in particular provides an online prediction system and method for powder injection molding injection blank density.

Background technique:

Powder injection molding technology is a near-net-shaping technology for parts formed by combining traditional powder metallurgy technology and modern plastic injection molding technology. It can use mold injection molding blanks and quickly manufacture high-density, high-precision, shape Complex structural parts, because of its unique advantages, are known as "the most popular part forming technology today". However, the defects generated in the injection molding process, that is, the uneven density distribution of the injection molding, has always been one of the main problems that plague people. At present, in the production of powder injection molding, the judgment of the defects of the injection blank is generally to cut the injection blank with qualified appearance quality and observe whether there are pores, cracks, sandwiches, etc. on the section. This manual inspection method is not only time-consuming and laborious, but also cannot be detected. Therefore, it is usually impossible to judge the existence of defects comprehensively and accurately, and its accuracy depends more on the experience of the operator.

The transformation of the production process of powder injection molding from manual mechanization to automatic intelligence is an important development direction in the future. To realize the automatic intelligent control of powder injection molding, a certain technology must be applied in the injection molding stage to monitor the density distribution of the injection blank. In this way, a judgment can be made on whether there are internal defects in the injection billet, so as to provide a subsequent intelligent adjustment plan for parameters. Introducing the online prediction technology of injection billet density into powder injection molding can predict the quality of the injection billet in advance, eliminating the tediousness and error caused by manual judgment. There are no research reports in this area at home and abroad.

Invention content:

The purpose of the present invention is to establish an online prediction system and method for powder injection molding injection molding density, which can avoid the tediousness and errors caused by manual judgment, and make online intelligent control of powder injection molding parameters possible.

The purpose of the present invention is achieved through the following technical solutions:

An online prediction system for the density of a powder injection molding blank, characterized in that the system includes an injection molding machine, an industrial CT machine, an image processing system, a sensor network system, and an artificial neural network system; wherein:

The injection molding machine is used to make the injected powder into a green body sample;

The industrial CT machine is used to scan the green body sample to generate a DR projection map of the sample;

The image processing system is used to obtain the overall gray value H and local gray value L _i of the green body sample;

The sensor network system is used to monitor the temperature value T and the pressure value P of each point inside the mold;

The artificial neural network system is used to establish the above-mentioned non-linear mapping relationship between T, P and H, L _i , and give the predicted values of H, L _i according to T, P values.

Further, in the above technical solution, the artificial neural network system is composed of a man-machine interface, a data processing system, and an artificial neural network; the man-machine interface is used for information exchange between the system and the user; the data processing system uses The T, P data collected by the sensor network system and the H and _Li values detected by the industrial CT machine are normalized and processed into data that can be recognized by the artificial neural network; the artificial neural network is used to establish the internal parameters of the mold T, The nonlinear mapping relationship between P and the gray distribution H and L _i of the injection green body, and the predicted gray value is given.

Further, in the above technical solution, in the sensor network system, sensors are installed in a direction parallel to the fluid flow phase to detect the temperature value T and pressure value P of each point.

An online method for predicting the density of a powder injection molding billet, characterized in that it specifically includes the following steps:

(1) Feed the uniform powder into the injection molding machine, inject with any set of process parameters, and form a green body sample;

(2) When the injection process is completed, the sensor network automatically detects the temperature value T and pressure value P of each point inside the injection mold;

(3) The above T and P are automatically sent to the artificial neural network system, and the neural network system automatically gives the predicted values of the overall gray level H and local gray level L _i of the injected green body sample.

Further, in the step (1), the process parameters include injection pressure, injection speed, and injection temperature.

Further, in the described step (3), the artificial neural network system prediction specific method is:

The T and P data detected by the sensor network are used as input layer data, and the overall gray value H and local gray value _Li of the green body sample are used as output layer data, and both the hidden layer and the output layer of the artificial neural network use The hyperbolic tangent transfer function is sent to the untrained artificial neural network after data normalization processing, and it is trained, thereby establishing the nonlinear mapping relationship between the internal parameters of the mold and the density distribution of the injection body; to the trained artificial neural network The network can input a set of T and P values arbitrarily, and the trained artificial neural network system will automatically give H and _Li values.

Further, the data normalization method is specifically as follows:

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; Dm</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.95</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; 1.05</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.95</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; min</span>}}$

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; T\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.95</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; T</span>}\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; 1.05</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; T</span>}\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.95</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; T</span>}\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

分别为第m个样本神经网络输入、输出归一化值；Dm、Tm分别为第m个样本的输入、输出标定值；Dmin、Tmin、Dmax和Tmax分别为不同输入输出的最小、最大标定值。

They are the normalized values of the input and output of the neural network of the mth sample; Dm and Tm are the calibration values of the input and output of the mth sample respectively; Dmin, Tmin, Dmax and Tmax are the minimum and maximum calibration values of different input and output respectively .

The advantages of the present invention are:

1. The automatic prediction of the density distribution of the injection blank saves a lot of tedious labor brought by the traditional manual operation method and improves the efficiency.

2. The testing process does not destroy the sample, and qualified samples can still be used, saving costs.

3. The quality of injection products can be monitored in real time, so that the quality problems of injection billets can be found in time.

Description of drawings

Fig. 1 is a structural block diagram of an online prediction system for powder injection molding injection molding density according to the present invention.

Fig. 2 is a flow chart of an online prediction system for powder injection molding injection molding density of the present invention.

Fig. 3 is a schematic diagram of the area division of the injection body in the present invention.

Detailed ways

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and implementation examples.

As shown in Fig. 1, it is a structural block diagram of an online prediction system for powder injection molding injection molding density according to the present invention. The block diagram shown includes two parts, ① is the training structure diagram of the artificial neural network, and ② is the application structure diagram of the trained artificial neural network. As shown in Figure 1, the system includes an injection molding machine, an industrial CT machine, an image processing system, a sensor network system, and an artificial neural network system; among them:

The injection molding machine is used to make the injected powder into a green body sample;

The industrial CT machine is used to scan the green body sample to generate a DR projection map of the sample;

The image processing system is used to obtain the overall gray value H and local gray value L _i of the green body sample;

The sensor network system is used to monitor the temperature value T and the pressure value P of each point inside the mold;

The artificial neural network system is used to establish the nonlinear mapping relationship between the above-mentioned T, P and H, _Li , and provides the predicted value of H, _Li according to T, P values; the artificial neural network system is composed of man-machine interface, data processing system, and artificial neural network; the human-machine interface is used for information exchange between the system and the user; the data processing system is used to detect the T and P data collected by the sensor network system and the industrial CT machine The normalized processing of the H and _L values obtained is data that can be identified by the artificial neural network; the artificial neural network is used to establish the nonlinear mapping relationship between the mold internal parameters T, P and the gray distribution H and _L of the injection body , and give the predicted gray value.

Fig. 2 is a method flow chart of an online prediction system for powder injection molding injection molding density according to the present invention. As shown in Figure 2, first feed the uniform powder into the injection molding machine to form an injection blank sample; when the injection is completed, the sensor network system monitors the temperature value T and pressure value P of each point inside the mold; the said After the sample is released from the mold, it is placed on the conveyor belt at a fixed angle and transported to the industrial CT monitoring equipment. The fixed angle is the angle between the central section of the sample and the detector panel; when the sample runs to the center of the CT equipment , the CT machine scans to obtain the DR projection map of the sample; transmit the DR projection map to the image processing system, select the entire area of the DR map of the sample, and obtain the overall gray value H of the sample; the DR projection map of the selected sample For each local area in the figure, the average gray value L _i of each local area of the sample is obtained. T and P are used as the input layer data of the artificial neural network, and H and _Li are used as the output layer data of the artificial neural network. Normalization processing is carried out in the data processing system, and then input to the untrained artificial neural network for training; after the training is completed, a trained artificial neural network system is obtained, and the trained artificial neural network system already has the ability of gray scale prediction. When it is necessary to predict the density of the injection blank under a certain injection parameter, it is only necessary to obtain the temperature value T _i and pressure value P _i of each point inside the mold when the injection is completed through the sensor network detection, and then input it to the trained artificial nerve Network system, the system will automatically give the overall gray value distribution and local gray value distribution of the injection blank.

Implementation example:

Choose 316L stainless steel powder, the binder is 79% paraffin + 20% high-density polyethylene + 1% stearic acid, and the powder loading is 53%. The powder and binder were mixed at 140°C-150°C for 1.5h to obtain uniform feeding.

A cuboid sample is injected on an injection machine, and the mold size is 28.3mm×20mm×6mm. Use the following series of parameter combinations for injection, the mold temperature is kept at 300K; the injection temperature is between 420K and 450K, and an injection point is taken every 5K; the injection rate is between 60cm ³ /S and 90cm ³ /S, every 2cm ³ /S interval to take an injection point. A cuboid sample was injected under each set of parameters, and a total of 1×2×16=32 injections were made to obtain 32 injection blank samples. At the same time, the temperature value T _i and the pressure value P _i of each point inside the mold are monitored through the sensor network when each injection is completed.

Put each injection blank sample into the industrial CT machine, and the section where the length and width of the sample are located is parallel to the detection panel. Scan with a CT machine to obtain the DR projection map of each sample. When the CT machine scans, the X-ray tube voltage is 120kV, the tube current is 225μA, and the magnification of the projection map is 10 times.

Input the DR map of each sample into the image processing software to obtain the overall gray value H _i and local gray value L _ij information of each sample.

Each group of T _i , P _i and H _i , L _ij is input to the data processing system through the man-machine interface for normalization processing, and then sent to the untrained artificial neural network for training, where T _i and P _i are the input Layer data, H _i , L _ij are output layer data. From this, the nonlinear relationship model between T _i , P _i and H _i , L _ij is established, and a mature artificial neural network density system is obtained.

The following two sets of injection parameters were selected to test the prediction effect of the online prediction system for the density of the powder injection molding injection molding to verify its accuracy.

Example 1: A set of parameters such as mold temperature 300K, injection temperature 420K, and injection rate 63cm ³ /S are used for injection, and the temperature and pressure values of each point inside the mold are obtained by monitoring the sensor network system as follows: (T1, P1)=(419.5 , 81.6); (T2, P2) = (421.3, 83.5); (T3, P3) = (419.5, 81.6); (T4, P4) = (420.0, 82.8); (T5, P5) = (422.5, 84.3 ); (T6, P6) = (420.0, 82.8); (T7, P7) = (418.6, 80.6); (T8, P8) = (420.2, 83.0); (T9, P9) = (418.6, 80.6)

The data obtained from the above monitoring is normalized through the man-machine interface and then input to the artificial neural network system for grayscale prediction, and at the same time, the grayscale detection of the same sample is carried out through the industrial CT machine. The predicted and detected values are listed in the table below:

h L1 L2 L3 L4 L5 L6 L7 L8 L9 Predictive value 690 708 700 708 688 680 688 684 672 684 Measurements 690 707 698 708 688 680 687 684 673 685

It can be seen that the predicted value of the artificial neural network system is very close to the detected value of industrial CT.

Example 2: A set of parameters such as mold temperature 300K, injection temperature 422K, and injection rate 75cm ³ /S are used for injection, and the temperature and pressure values of each point inside the mold are obtained by monitoring the sensor network system as follows: (T1, P1) = (421.6 , 82.2); (T2, P2) = (423.5, 84.3); (T3, P3) = (421.5, 82.2); (T4, P4) = (422.6, 83.5); (T5, P5) = (424.3, 84.8 ); (T6, P6) = (422.5, 83.6); (T7, P7) = (419.8, 81.9); (T8, P8) = (421.3, 83.5); (T9, P9) = (419.8, 81.8)

The data obtained from the above monitoring is normalized through the man-machine interface and then input to the artificial neural network system for grayscale prediction, and at the same time, the grayscale detection of the same sample is carried out through the industrial CT machine. The predicted and detected values are listed in the table below:

h L1 L2 L3 L4 L5 L6 L7 L8 L9 Predictive value 653 670 662 670 650 642 650 646 638 646 Measurements 653 670 661 671 651 643 650 647 636 648

It can be seen that the predicted value of the artificial neural network system is very close to the detected value of industrial CT.

Claims (7)

1. a powder injection forming base density on-line prediction system is characterized in that this system comprises injection machine, Industrial CT Machine, image processing system, sensor network system, artificial neural network system; Wherein:

Said injection machine is used for the powder that injects is processed the base substrate sample;

Said Industrial CT Machine is used to scan the base substrate sample, generates the DR perspective view of sample;

Said image processing system is used to obtain base substrate sample overall gray value H and local gray-value L _i

Said sensor network system is used to monitor temperature value T, the pressure value P of mould inside each point;

Said artificial neural network system is used to set up above-mentioned T, P and H, L _iNonlinear Mapping relation, and provide H, L according to T, P value _iPredicted value.

2. a kind of powder injection forming base density on-line prediction according to claim 1 system, it is characterized in that: said artificial neural network system is made up of man-machine interface, data handling system, artificial neural network; Said man-machine interface is used for the information exchange between system and the user; Said data handling system is used for T, P data and the detected H of Industrial CT Machine, L that sensor network system is collected _iValue normalization is treated to artificial neural network can recognition data; Said artificial neural network is used to set up mould inside parameter T, P and injection base substrate intensity profile H, L _iNonlinear Mapping relation, and provide the gray scale predicted value.

3. a kind of powder injection forming base density on-line prediction according to claim 1 system, it is characterized in that: in the said sensor network system, sensor installation detects the temperature value T and the pressure value P of each point on the direction parallel with fluid flowing phase.

4. powder injection forming base density on-line prediction method is characterized in that specifically may further comprise the steps:

4.1 the uniform powder feeding is sent in the injection machine, inject with any one group of technological parameter, form the base substrate sample;

4.2 accomplish constantly at injection process, sensor network detects the temperature value T and the pressure value P of the inner each point of injection molding automatically;

4.3 give the artificial neural network system with above-mentioned T, P automatic transport, nerve network system provides injection base substrate sample overall intensity H and local gray level L automatically _iPredicted value.

5. a kind of powder injection forming base density on-line prediction method according to claim 4 is characterized in that in the said step 4.1, technological parameter comprises injection pressure, injection speed, injection temperature.

6. a kind of powder injection forming base density on-line prediction method according to claim 4 is characterized in that in the said step 4.3, the artificial neural network system predicts that concrete grammar is:

With the detected T of said sensor network, P data as the input layer data, with base substrate sample overall gray value H and local gray-value L _iAs the output layer data; Artificial neural network hidden layer and output layer all adopt the tanh transfer function; Do not instruct artificial neural network through being sent to after the data normalization processing, and it is done training, set up the mould inside parameter thus and inject the Nonlinear Mapping relation that blank density distributes; To instructing artificial neural network one group of T of input, P value arbitrarily, instructed the artificial neural network system and can provide H, L automatically _iValue.

7. a kind of powder injection forming base density on-line prediction method according to claim 6 is characterized in that:

Said data normalization mode is specially:

$\stackrel{\&OverBar;}{D}=\frac{\mathrm{Dm}-0.95*D\min }{1.05*D\max -0.95*D\min }$

$\stackrel{\&OverBar;}{T}=\frac{\mathrm{Tm}-0.95*T\min }{1.05*T\max -0.95*T\min }---\left(1\right)$

is respectively m sample neutral net input, output normalized value; Dm, Tm are respectively input, the output calibration value of m sample; Dmin, Tmin, Dmax and Tmax are respectively the minimum of different input and output, maximum calibration value.

Images