Mass production electric control PCBA aging test anomaly detection system and detection method thereof

Abstract

The present application relates to mass production of electric control PCBA aging test abnormality detection system and detection method, industrial control unit contains program-controlled power supply, relay channel switch unit and ammeter, aging test room is arranged with multiple aging test racks, each aging test rack is equipped with multiple aging test boxes; Industrial computer is connected with multiple program-controlled power supply through signal control bus, controls the output of specific DC voltage and current, connects ammeter to obtain the current value of channel in real time, connects temperature control unit and humidity control unit, issues temperature and humidity control instructions and obtains temperature and humidity measurement values in real time, temperature control unit is connected with heating heater and temperature sensor group in aging test room, controls the temperature change in aging test room, humidity control unit is connected with humidifier and humidity sensor group in aging test room, controls the humidity change in aging test room; Industrial control unit is connected with aging test box, supplies power to PCBA to be tested and monitors in real time. Greatly improve the test efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to a mass production electric control PCBA aging test anomaly detection system and a detection method thereof. BACKGROUND

[0002] At present, in recent years, the industrial internet has developed rapidly, various digital automatic factory applications have emerged in an endless stream, and various digital and automatic applications have also appeared. Burn-in test is a method of screening out early failures from a batch of devices to be tested before they are put into actual use by customers.

[0003] The PCBA (Printed Circuit Board Assembly) aging test used on the production line includes temperature cycling or humidity cycling according to specific requirements. One test method is to place the aging products into an aging room for high-low temperature and high-low humidity reliability testing without power on. Another test method is to place the products into an aging room for aging test with power on. After the completion of the two test methods, various function tests are performed to screen out the problematic PCBA products.

[0004] The existing technology and equipment encounter problems when performing power-on aging test on a large scale of mass production PCBA in the factory. Due to the small number and high price of PCBA function test machines, it is not possible to individually place each PCBA into a function test machine for online aging test during mass production. A group of PCBA to be tested needs to be placed into a specific aging test room, and the temperature and humidity cycle control process needs to be set before the aging test. On the one hand, overheating, short circuit and other failures may occur during the aging test, which may cause mutual influence of the tested PCBA and lead to associated losses. At the same time, EHS (Environment, Health, Safety) accidents are also prone to occur. On the other hand, due to the operation errors of production line employees, including but not limited to incorrect PCBA insertion quantity and improper connection of various connection interfaces, the aging test may be abnormal and the results may be unreliable, which affects the aging test efficiency. The existing technology and system cannot meet the actual needs of mass production PCBA aging test.

[0005] In summary, it is necessary to design a mass production PCBA aging test system based on different PCBA test types. In the case of small data and non-supervised state, i.e. without a large amount of data and without manual setting of various threshold values and human intervention, it is necessary to determine whether the tested PCBA is abnormal in real time and to alarm in time, reduce EHS accidents and test losses, and improve the aging test efficiency, which has become a problem to be solved. SUMMARY

[0006] The application aims to overcome the shortcomings of the prior art and provide a mass production electric control PCBA aging test abnormality detection system and a detection method thereof.

[0007] The application aims to overcome the shortcomings of the prior art and provide a mass production electric control PCBA aging test abnormality detection system and a detection method thereof.

[0008] The mass production electric control PCBA aging test abnormality detection system comprises a display terminal, an industrial control unit and an aging test room.

[0009] Further, the mass production electric control PCBA aging test abnormality detection system, wherein the aging test box comprises a test base plate and a base plate connecting card slot, the PCBA under test is connected to the test base plate through the base plate connecting card slot, and the test base plate supplies power to the PCBA under test.

[0010] Further, the mass production electric control PCBA aging test abnormality detection system, wherein the industrial control computer is connected to the programmable power supply, the relay channel switch unit, the ammeter, the temperature control unit and the humidity control unit through a signal control bus, the programmable power supply is connected to the relay channel switch unit through a power supply power line, the power supply output is controlled through the relay channel switch unit, the relay channel switch unit is connected to the ammeter through the power supply power line to acquire current data in real time, the aging test box is connected through the power supply power line to switch current and voltage output and acquire current data of power supply in real time, and safe operation and data acquisition of the aging test are realized.

[0011] Further, the mass production electric control PCBA aging test abnormality detection system, wherein the temperature control unit of the industrial control unit is connected to the heating electric heater and the temperature sensor group in the aging test room through a signal control bus, the humidity control unit is connected to the humidifier and the humidity sensor group in the aging test room through a signal control bus, and the industrial control unit is connected to the aging test box in the aging test room through a power supply power line.

[0012] Further, the mass production electric control PCBA aging test abnormality detection system, wherein the signal control bus is a modbus bus based on RS485 or a deviceNet bus based on network.

[0013] Further, the mass production electric control PCBA aging test abnormality detection system, wherein the power supply power line is a double-core shielded power line.

[0014] The mass production electric control PCBA aging test abnormality detection method comprises the following steps:

[0015] (1) Connect the aging test box with the industrial control unit, scan the code PCBA and the corresponding aging test box to bind; the ammeter of the industrial control unit samples the current data of each aging test box at equal intervals, continuously samples a complete aging test period, and after no abnormality occurs, it is regarded as a group of normal vector data;

[0016] (2) Initialize time series data, segment sample and mean process the continuous data;

[0017] (3) Improve the VAE model, introduce the LSTM recurrent neural network, take the sliced time series data as the input of model training, generate a specific normal working model based on the aging test data set on the deep learning workstation, and store it in the hierarchical data format file HDF5;

[0018] (4) Convert the model generated on the deep learning workstation into a C / C++ file-based model, and compile and link the aging detection control system with the abnormality detection model;

[0019] (5) Deploy the aging test abnormality detection model on the industrial control computer to run, collect the time series-based current data in real time, calculate the judgment function Rt of the model, and if the result is greater than the threshold, make a predictive alarm.

[0020] Further, the mass production electric control PCBA aging test abnormality detection method comprises the following steps:

[0021] (1) Connect the aging test box with the industrial control unit, scan the code PCBA and the corresponding aging test box to bind; the ammeter of the industrial control unit samples the current data of each aging test box at equal intervals, continuously samples a complete aging test period, and after no abnormality occurs, it is regarded as a group of normal vector data;

[0022] According to the established process, power on and perform aging test, the ammeter of the industrial control unit samples the current data of each aging test box at equal intervals of 1s-10s, continuously samples a complete aging test period of 24 hours-120 hours;

[0023] Put all the PCBA that passed the test into the FT function test machine for testing. After passing the test without any abnormalities, use the above time series data as a set of normal vector data.

[0024] (2) The collected current data is averaged according to the number of PCBA in each aging test box, i.e. the total current detected by each aging test box is divided by the number of PCBA in the test box;

[0025] Initialize the time series data, and perform mean value processing every 5 data. With a 1s sampling interval period, 5s mean value processing is performed every 10 minutes as a time window. Each time window has 120 collected and averaged current data, forming a 120-dimensional vector as a window time series data.

[0026] The aging test room can put in 10 test racks, each test rack put in 6 test boxes, full load aging test can obtain 60 groups of long time series data, PCBA through aging test, based on time series current data has regularity, multiple data according to the same every 10 minutes as a time window for slicing, as the data input of VAE model;

[0027] (3) Construct VAE model, VAE model contains encoder and decoder, X1, X2, X3…X n The real vector data input from the outside is calculated by mean and variance to obtain the mean and variance of different normal distribution models. After reparameterization calculation, the latent variable z is subject to normal distribution N(0, I). After sampling, the latent variables Z1, Z2, Z3…Z n , after the decoder, the generated reconstruction data X'1, X'2, X'3…X' n , by minimizing the loss function￡, the encoder and decoder are constructed to obtain the aging anomaly detection model based on current data:

[0028]

[0029] Е x ~ p(x) [f(x)] represents the expectation of f(x), where the distribution of x is p(x);

[0030] KL(p(x)║q(x)) represents the KL divergence of two distributions;

[0031] The symbol represents the loss function;

[0032] An improved VAE model based on time series data is constructed. The original time window input model is encoded using the VAE encoder to obtain a low-dimensional set of latent variable vectors, which are then embedded into an LSTM to predict the embedded input of the next window. After the LSTM model is used to operate, the embedded output is predicted, and then the VAE decoder is used to decode to obtain the reconstructed time series data window. The prediction error of the reconstructed time series data is used as the anomaly judgment criterion.

[0033] A local window containing p consecutive readings is used as input. An encoder estimates the q-dimensional low-dimensional embedding, and a decoder reconstructs the original window. A scrolling window is generated from the training data to train the VAE model. t =[x t-p+1 ,…,x t The window ends at the end of the time series window t, the continuous reading p is 120, the range is 60 to 180 dimensions, and the encoder output latent variable dimension q is 20 dimensions, the range is 10 to 30 dimensions;

[0034] Introducing the LSTM (Long Short-Term Memory) recurrent neural network model to process time series data, using W... t =[w t-(k-1)*p ,w t-(k-2)*p ,…,w t ] represents a time series, which is embedded in the VAE of k non-overlapping window sequences by a VAE encoder; W t After transformation by the VAE encoder, the corresponding embedded input is used to update and optimize the VAE model parameters using normal time series data from aging tests by minimizing the loss function ε, i.e., formula (1). After optimizing the VAE model, the encoder from the trained VAE model is used to estimate all embedded sequences E in the training dataset. t To train the LSTM model, the LSTM model takes sequence E. t The first k-1 embedding in the matrix, and predict the next embedding, i.e.:

[0035]

[0036] This represents the embedded input of the (k-1)th window of the time series window;

[0037] This represents the k-th embedded output predicted by the LSTM model after the time series window is processed.

[0038] LSTM stands for Recurrent Neural Network Module;

[0039] The LSTM model algorithm can directly call the keras.layers.LSTM in the open source module keras;

[0040] The LSTM model parameters are optimized by minimizing the embedded prediction error, i.e.

[0041] The working model based on the aging test data set is generated on the deep learning workstation by using the above method with python programming, and the hierarchical data format file HDF5 is output after training is completed;

[0042] (4) The model generated on the deep learning workstation is converted into a model based on C / C++ files, and the anomaly detection model is compiled and linked into the aging detection control system;

[0043] (5) The aging test anomaly detection model is deployed on the industrial computer to run, real-time acquisition of time series based current data, through VAE encoder conversion, VAE hidden variable corresponding to the next window predicted by LSTM, according to the hidden variable, using the trained VAE for reconstruction;

[0044]

[0045] represents the kth embedded output predicted by the LSTM model for the time series window;

[0046] represents the reconstructed time series;

[0047] Decoder represents the VAE decoder;

[0048] For the reconstructed window, a function R t is defined to evaluate whether the test is abnormal by calculating the w t prediction error;

[0049]

[0050] w t-(k-i)*p represents the original input time series;

[0051] represents the reconstructed time series;

[0052] R t represents the prediction error;

[0053] A threshold is set, and the calculation is carried out according to the above formula (4), if the result is greater than the set threshold, it is determined that the aging anomaly occurs, after the aging test time window is ended, the predictive alarm is made.

[0054] Compared with the prior art, the present application has remarkable advantages and beneficial effects, which are embodied in the following aspects:

[0055] ①The present application integrates various control hardware and sensors to realize digital management of the whole process of mass production PCBA aging test, increase the traceability of the test process, and improve the production efficiency of the test;

[0056] ②The hardware structure is simple and has strong universality;The aging test system provides power supply and collects corresponding voltage and current data, and can be compatible with mass aging test of different kinds of PCBA by replacing the test board in the aging box, reducing subsequent development investment;Without additional hardware investment cost, the aging abnormality detection can be completed by adding a functional test equipment module, the predictive alarm reduces the probability of EHS accidents, and at the same time, the test inefficiency caused by the operation mistakes of the staff such as improper plugging or missing plugging can be found in time, improving the test efficiency;

[0057] ③VAE algorithm in deep learning algorithm is adopted, the advantage of VAE is to provide probability measurement instead of reconstruction error as abnormal score, which is called reconstruction probability, the algorithm reconstruction probability is a probability measurement considering the variability of variable distribution, compared with the reconstruction error used in automatic encoder AE and principal component PCA based abnormality detection method, the algorithm has more principle and objectivity in the application of aging test abnormality detection field;

[0058] ④Improved VAE model method, on the basis of VAE detecting abnormality, the input time series is sliced, after encoding by VAE encoder, a regression prediction model LSTM is introduced in the hidden variable layer, which can make real-time predictive alarm without exceeding the time window length, reduce the probability of EHS accidents, and improve the test efficiency;

[0059] ⑤Based on small data and unsupervised training data model, that is, without the large amount of data training required by deep learning algorithm to obtain practical algorithm model, the current data is sampled at an interval of 1 minute, 48 hours of aging test period, and about 100 groups of data can complete the training of the algorithm model;The model is constructed in an unsupervised manner, that is, without adding various labels manually, which can reduce the personnel development and deployment cost, and realizes rapid development and deployment for different varieties of mass production PCBA;

[0060] ⑥It is used for managing the whole process of mass production PCBA aging test process, and detecting test abnormality in real time, achieving the expected aging test abnormality predictive alarm, effectively reducing the loss of the factory, and improving the efficiency of the aging test;At the same time, it is connected with the factory MES system to realize the whole process digital management of PCBA production test.

[0061] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0063] Fig. 1 Fig. 1 is a structural schematic diagram of the detection system of the present application;

[0064] Fig. 2 Fig. 3 is a structural schematic diagram of the industrial control unit;

[0065] Fig. 3 Fig. 5 is a structural schematic diagram of the aging test box;

[0066] Fig. 4 Fig. 7 is a flow schematic diagram of the detection method of the present application;

[0067] Fig. 5 Fig. 9 is a schematic diagram of training the aging anomaly detection variational autoencoder model based on current data;

[0068] Fig. 6 Fig. 11 is a schematic diagram of the improved VAE model based on time series. DETAILED DESCRIPTION

[0069] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0070] It should be noted that similar reference numerals and letters refer to like items in the several figures of the drawings, and that, as such, once an item is defined in one figure, it should not require further defining and explaining in subsequent figures. Also, in the description of the present application, orientation terms and ordinal terms, etc. are used only to distinguish one part from another part, and should not be understood to indicate or imply relative importance.

[0071] As shown in Figs. 1-3 The mass production electric control PCBA aging test anomaly detection system includes a display terminal 1, an industrial control unit 2, and an aging test room 3. The industrial control unit 2 includes a programmed power supply 21, a relay channel switch unit 22, an ammeter 23, a temperature control unit 24, a humidity control unit 25, and an industrial control computer 26. The aging test room 3 is arranged with multiple aging test racks 32, each of which is equipped with multiple aging test boxes 31 for loading the PCBA to be tested. The industrial control computer 26 is connected to the multiple programmed power supplies 21 through a signal control bus 27 to control the output of specific DC voltage and current, connected to the ammeter 23 to obtain the current value of the channel in real time, connected to the temperature control unit 24 and the humidity control unit 25 to issue temperature and humidity control instructions and obtain temperature and humidity measurement values in real time. The temperature control unit 24 of the industrial control unit 2 is connected to the heating element and temperature sensor group in the aging test room 3 through the signal control bus 27 to control the temperature change in the aging test room 3 through a PID algorithm in a closed loop. The humidity control unit 25 is connected to the humidifier and humidity sensor group in the aging test room through the signal control bus 27 to control the humidity change in the aging test room 3 through a PID algorithm in a closed loop. The industrial control unit 2 is connected to the aging test boxes in the aging test room through a power supply power line 28. The industrial control unit 2 is connected to the DC power supply line group to supply power to the PCBA to be tested and monitor in real time.

[0072] The aging test box 31 includes a test base plate 311 and a base plate connection card slot 313. The PCBA to be tested 312 is connected to the test base plate 311 through the base plate connection card slot 313. The test base plate 311 supplies power to the PCBA to be tested 312.

[0073] The industrial computer 26 is connected with the program-controlled power supply 21, the relay channel switch unit 22, the ammeter 23, the temperature control unit 24 and the humidity control unit 25 through the signal control bus 27 respectively, the program-controlled power supply 21 is connected with the relay channel switch unit 22 through the power supply power line 28, the power output is controlled through the relay channel switch unit 22, the relay channel switch unit 22 is connected with the ammeter 23 through the power supply power line 28, the current data is acquired in real time, the aging test box 31 is connected through the power supply power line 28, the current data of the power supply is acquired in real time through the switching of current and voltage output, the safe operation and data acquisition of the aging test are realized. The signal control bus 27 is a modbus bus based on RS485 or a deviceNet bus based on network. The power supply power line 28 is a double-core power line with shielding.

[0074] The display terminal 1 comprises a display with a touch screen, a keyboard and a mouse; the aging test room is composed of heat preservation library board compartments, and is configured with heating electric heaters, humidifiers, trays, sliding doors, circulating fans, exhaust fans and the like, so as to simulate a high temperature and harsh environment, so that the product is connected with a simulated load and operated in the environment; the aging test box is used for loading the PCBA to be tested and providing power supply and load; the aging test frame is used for loading the aging test box, and the bottom is provided with a roller device, so that a group of aging test frames can be conveniently transported into the aging test room. The industrial control unit is connected with each unit in the aging test room 3 through the signal control bus 27, and each aging test box 31 on the aging test frame 32 is powered through the power supply power line 28.

[0075] The program-controlled power supply 21 is a direct current program-controlled power supply, such as DP3000 of Puyuan Jingdian; the relay channel switch unit 22 is a relay control module group, such as QJ-32RG-T1Z-24V of Qingjun Electronics; the ammeter is DW-81-B / C of Chuanghong Instrument; the temperature control unit 24 has PID feedback control, such as AT900-R-A1-RS-P2 of Chuanghong Instrument; the humidity control unit 25 has PID feedback control, such as WHD46 of Ankerui Electrical Co., Ltd.; the industrial computer 26 is a general-purpose computer, which has a plurality of serial ports and network interfaces, such as IBOX-105V2-2L4C of Tuolang Industrial Computer.

[0076] According to different PCBA test board types, the bottom plate in different aging test boxes can be replaced. Due to the fixedness of the temperature and humidity cycle process in the aging test, the current and other data supplied by the industrial control unit to the aging test box are collected in real time. In the state of small data and non-supervision, that is, without manual setting of various threshold values and human intervention, a deep learning algorithm is used for real-time calculation and analysis of whether there is a test anomaly in each PCBA in the aging test box, to provide predictive alarm for large-scale aging test. If there is an anomaly, the power supply to the corresponding aging test box is cut off in time, thereby reducing the loss. It can also be an operation error of the employee, PCBA missing insertion or insertion out of position, resulting in abnormal current based on time series data, thereby issuing a predictive alarm in real time and making corrections in time, thereby improving the overall aging test efficiency.

[0077] The mass production of electric control PCBA aging test anomaly detection process is as follows: Fig. 4 The specific steps are as follows:

[0078] (1) Connect the aging test box to the industrial control unit, scan the surface two-dimensional code of the PCBA to be tested and the two-dimensional code of the aging test box with the code gun, and bind with the corresponding aging test box;

[0079] According to the established process, power on and perform aging test. The current meter of the industrial control unit samples the current data of each aging test box at an equal interval of 1s-10s, and continuously samples for a complete aging test period of 24-120 hours;

[0080] Put all the tested PCBA into the FT function test machine for testing. After passing the test without exception, the time series data is taken as a group of normal vector data;

[0081] (2) Process the collected current data according to the number of PCBA in each aging test box, that is, divide the total current detected by each aging test box by the number of PCBA in the test box;

[0082] Initialize the time series data, do mean value processing every 5 data, 1s sampling interval, 5s mean value processing, and every 10 minutes as a time window, 120 collected and mean value processed current data in each time window, form a 120-dimensional vector as a window time series data;

[0083] The aging test room can put in 10 test racks, each test rack puts in 6 test boxes, and full load aging test can obtain 60 groups of long time series data. After the PCBA passes the aging test, the time series current data has regularity. Slice the data according to the same time window of every 10 minutes as the data input of the VAE model;

[0084] (3) Constructing a VAE model, such as Fig. 5 The VAE model comprises an encoder and a decoder, X1, X2, X3…X n The real vector data inputted externally is subjected to mean variance calculation to obtain the mean and variance of different normal distribution models, and is subjected to reparameterization calculation to make the latent variable z subject to normal distribution N(0, I), and is sampled to obtain latent variables Z1, Z2, Z3…Z n After the decoder, the generated reconstructed data X'1, X'2, X'3…X' n The encoder and the decoder are constructed by minimizing the loss function £, and an aging anomaly detection model based on current data is obtained:

[0085]

[0086] E x ~ p(x) [f(x)] represents the expectation of f(x), wherein the distribution of x is p(x);

[0087] KL(p(x)║q(x)) represents the KL divergence of two distributions;

[0088] The symbol of the loss function is represented;

[0089] An improved VAE model based on time series data is constructed, such as Fig. 6 The original time window input model is encoded by the encoder of the VAE to obtain a low-dimensional latent variable vector set, which is embedded into the LSTM to predict the embedded input of the next window. After the operation of the LSTM model, the embedded output is predicted, and then the VAE decoder is used for decoding to obtain the reconstructed time series data window. The prediction error of the reconstructed time series data is used as an abnormality judgment standard;

[0090] The following is explained: the local window of p consecutive readings is taken as input, the q-dimensional low-dimensional embedding is estimated by the encoder, and the original window is reconstructed by the decoder; for training the VAE model, the rolling window is generated from the training data; w t =[x t-p+1 ,…,x t ] represents that the window ends at time series window t, the number of consecutive readings p is 120, the range is 60-180 dimensions, the dimension of the encoder output latent variable q is 20, and the range is 10-30 dimensions;

[0091] The LSTM (Long Short-Term Memory) recurrent neural network model is introduced to process time series data, and W t =[w t-(k-1)*p ,w t-(k-2)*p ,…,wt represents the time series, the VAE encoder is used to perform the embedding operation on the k non-overlapping window sequence of the VAE; represents W t After the conversion of the VAE encoder, the corresponding embedded input is obtained by using the normal time series data of the aging test, the loss function￡is minimized, that is, formula (1), the VAE model parameters are updated by using the stochastic gradient descent optimization algorithm, and after the optimization of the VAE model, all embedded sequences E t in the training data set are estimated by using the encoder from the trained VAE model; t To train the LSTM model, the LSTM model takes the first k-1 embedded sequences in the sequence E

[0092]

[0093] represents the embedded input of the k-1 window of the time series window;

[0094] represents the k embedded output predicted by the LSTM model for the time series window;

[0095] LSTM represents a recurrent neural network module;

[0096] The LSTM model algorithm can directly call the keras.layers.LSTM in the open source module keras;

[0097] The LSTM model parameters are optimized by minimizing the prediction error of the embedded sequence, that is,

[0098] The working model based on the aging test data set is generated on the deep learning workstation by using the above method and python programming, and the hierarchical data format file HDF5 is output after the training is completed;

[0099] (4) The model generated on the deep learning workstation is converted into a model based on C / C++ files, and third-party tools or manual rewriting are used to generate, such as h5dump, hDFview, h5cc / h5c++, and the anomaly detection model is compiled and linked into the aging detection control system;

[0100] (5) The aging test anomaly detection model is deployed on the industrial computer to run, real-time acquisition of time series-based current data is performed, the VAE encoder is converted, the VAE hidden variable corresponding to the LSTM predicted next window is obtained, and the trained VAE is used for reconstruction according to the hidden variable;

[0101]

[0102] represents the kth inner-embedded output predicted by the LSTM model for the time series window;

[0103] represents the reconstructed time series;

[0104] Decoder represents the VAE decoder;

[0105] For the reconstructed window, a function R t is defined to evaluate whether the test is abnormal by calculating the prediction error w t

[0106]

[0107] w t-(k-i)*p represents the original input time series;

[0108] represents the reconstructed time series;

[0109] R t represents the prediction error;

[0110] A threshold is set, and the calculation is performed according to the above formula (4), if the result is greater than the set threshold, it is determined that the aging abnormality occurs, and a predictive alarm is made after the aging test time window ends.

[0111] In summary, the application integrates various control hardware and sensors to realize digital management of the whole process of mass production PCBA aging test, increase the traceability of the test process, and improve the production efficiency of the test.

[0112] The hardware structure is simple and has strong universality; the aging test system provides power supply and collects corresponding voltage and current data, and can be compatible with mass aging test of different kinds of PCBA by replacing the test board in the aging box, reducing the subsequent development investment; without additional hardware investment cost, the aging abnormality detection and predictive alarm reduce the probability of EHS accidents, and can also timely find the invalid test caused by operation errors such as employee misplug or omission, thereby improving the test efficiency.

[0113] The VAE (Variational Auto-Encoder) algorithm in the deep learning algorithm is adopted, and the advantage of VAE is to provide a probability measurement instead of a reconstruction error as an abnormal score, which is called reconstruction probability. The reconstruction probability of the algorithm is a probability measurement considering the variability of the variable distribution, and compared with the reconstruction error used by the automatic encoder AE and the principal component PCA-based abnormality detection method to measure the abnormality, the reconstruction probability has more principle and objectivity in the application of the algorithm in the field of aging test abnormality detection.​

[0114] The conventional VAE trains the VAE model by using the complete aging test time series data as the income, although the test abnormal PCBA can be screened out, but cannot make predictive alarm in real time, needs to complete the entire test process to screen out the aging abnormal PCBA, there is limitation, the improved VAE model method, on the basis of VAE detecting abnormality, the input time series is sliced, after being encoded by the VAE encoder, a regression prediction model LSTM is introduced in the hidden variable layer, the real-time predictive alarm can be made within the time window length, the probability of EHS accident is reduced, and the test efficiency is improved.

[0115] The training data model based on small data and unsupervised is adopted, that is, the practical algorithm model can be obtained without the training of a large amount of data required by the deep learning algorithm, the current data is sampled at an interval of 1 minute, the algorithm model can be trained by using about 100 groups of data in a 48-hour aging test period; the model is constructed in an unsupervised manner, that is, various labels do not need to be added artificially, the personnel development and deployment cost can be reduced, and the rapid development and deployment of different varieties of mass-produced PCBA can be realized.

[0116] The method is used for managing the whole process of the aging test of the mass-produced PCBA, detecting test abnormalities in real time, achieving the expected predictive alarm of the aging test abnormality, effectively reducing the loss of the factory, and improving the efficiency of the aging test; meanwhile, the method is connected with the MES system of the factory, and realizes the digital management of the whole process of the PCBA production test.

[0117] The above merely describes the preferred embodiments of the present application and is not used to limit the present application, and for those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0118] The above merely describes the preferred embodiments of the present application and is not used to limit the present application, and for those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0119] It is to be noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component" can include a combination of two or more components. Additionally, the terms "comprise," "comprises," and "comprising," or any variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, unless otherwise indicated herein, the terms "first," "second," "third," etc., are used herein merely as labels, and are not intended to impose ordinal import.

Claims

1. A mass production electric control PCBA aging test anomaly detection method, characterized in that: The mass production electric control PCBA aging test abnormality detection system comprises a display terminal (1), an industrial control unit (2) and an aging test room (3). The industrial control unit (2) comprises a programmable power supply (21), a relay channel switch unit (22), an ammeter (23), a temperature control unit (24), a humidity control unit (25) and an industrial control computer (26). The aging test room (3) is arranged with a plurality of aging test racks (32), and each aging test rack is provided with a plurality of aging test boxes (31) for loading the PCBA to be tested. The industrial control computer (26) is connected with the plurality of programmable power supplies (21) through a signal control bus (27) to control the output of specific DC voltage and current, connected with the ammeter (23) to obtain the current value of the channel in real time, connected with the temperature control unit (24) and the humidity control unit (25) to issue the temperature and humidity control instructions and obtain the temperature and humidity measurement values in real time. The temperature control unit (24) is connected with the heating electric heater and the temperature sensor group in the aging test room (3) to control the temperature change in the aging test room (3). The humidity control unit (25) is connected with the humidifier and the humidity sensor group in the aging test room (3) to control the humidity change in the aging test room (3). The industrial control unit (2) is connected with the aging test box in the aging test room to supply power to the PCBA to be tested and monitor in real time. The steps include: (1) Connect the aging test box with the industrial control unit, and bind the PCBA with the corresponding aging test box by scanning the code. The ammeter of the industrial control unit samples the current data of each aging test box at equal intervals, and continuously samples for a complete aging test period. When no abnormality occurs, it is regarded as a group of normal vector data; (2) Initialize time series data, segment the continuous data and perform mean value processing; (3) Improve the VAE model, introduce the LSTM recurrent neural network, and use the sliced time series data as the input of the model training. A specific normal working model based on the aging test data set is generated on a deep learning workstation, and is stored in a hierarchical data format file HDF5; (4) Convert the model generated on the deep learning workstation into a C / C++ file-based model, and compile and link the aging detection control system with the abnormality detection model; (5) Deploy the aging test abnormality detection model on the industrial control computer to run, collect the time series-based current data in real time, calculate the judgment function Rt of the model, and make a predictive alarm if the result is greater than the threshold.

2. The mass production electric control PCBA aging test anomaly detection method according to claim 1, characterized in that: The steps include: (1) Connect the aging test box with the industrial control unit, and bind the PCBA with the corresponding aging test box by scanning the code on the surface of the PCBA to be tested and the aging test box with the code gun; According to the established process, power on and perform aging test. The ammeter of the industrial control unit samples the current data of each aging test box at equal intervals of 1s-10s, and continuously samples for a complete aging test period of 24 hours-120 hours; Put all the PCBAs that have passed the test into the FT function test machine for testing. After no abnormality is found, the time series data is regarded as a group of normal vector data. (2) The collected current data is averaged according to the number of PCBA in each aging test box, that is, the total current detected by each aging test box is divided by the number of PCBA in the test box; Initialize time series data, average every 5 data, sample interval period 1s, average every 5s, take every 10 minutes as a time window, each time window has 120 collected and averaged current data, form a 120-dimensional vector as a window of time series data; The aging test room can put in 10 test racks, each test rack puts in 6 test boxes, full load aging test can obtain 60 groups of long time series data, after PCBA passes the aging test, based on the regularity of time series current data, multiple data are sliced according to the same time window of 10 minutes as the data input of VAE model; (3) Construct a VAE model, the VAE model includes an encoder and a decoder, X1, X2, X3…X n The mean and variance of different normal distribution models are obtained by performing mean variance calculation on the real vector data input externally, and the latent variable z is subjected to normal distribution N(0, I) through reparameterization calculation, and the latent variables Z1, Z2, Z3…Z n are obtained after sampling. n After the decoder, the generated reconstructed data X'1, X'2, X'3…X' n are obtained. The encoder and the decoder are constructed by minimizing the loss function £, and an aging anomaly detection model based on current data is obtained: £ = E x～p(x) [Ez ~ p(z | x) [-ln q(x | z) + KL(P(z | x) || q(z))] Equation (1) Ε x～p(x) [f(x)] denotes the expectation of f(x), where x has distribution p(x); KL(p(x)║q(x)) represents the KL divergence of two distributions; £ represents the symbol of loss function; An improved VAE model based on time series data is constructed, the original time window input model is encoded by the encoder of VAE to obtain a low-dimensional hidden variable vector set, which is embedded into LSTM to predict the embedded input of the next window. After the operation of the LSTM model, the embedded output is predicted, and then the decoder of VAE is used for decoding to obtain the reconstructed time series data window. The prediction error of the reconstructed time series data is used as the abnormality judgment standard; Taking a local window of p consecutive readings as input, estimating a low-dimensional embedding of q dimensions by an encoder and reconstructing the original window by a decoder; for training the VAE model, generating rolling windows from the training data; w t = [x t-p+1 ,…,x t ] represents that the window ends at time series window t, the consecutive readings p are 120, the range is 60-180 dimensions, the encoder output hidden variable dimension q is 20 dimensions, and the range is 10-30 dimensions; Introduce LSTM recurrent neural network model to process time series data, W t = [w t-(k-1)*p ,w t-(k-2)*p ,…,w t ] represents a time series, and the VAE is operated on the k non-overlapping window sequence through the VAE encoder; W t After conversion by the VAE encoder, the corresponding embedded input is obtained, the normal time series data of the aging test is used, the loss function￡is minimized, that is, formula (1), the VAE model parameters are updated by using the stochastic gradient descent optimization algorithm, and after the VAE model is optimized, the encoder from the trained VAE model is used to estimate all embedded sequences E t in the training data set; for training the LSTM model, the first k-1 embedded in the sequence E t is taken, and the next embedded is predicted, that is: an embedded input representing a k-1th window of the time series window; denotes the kth embedded output predicted by the LSTM model for the time series window; LSTM represents the recurrent neural network module; The LSTM model algorithm can directly call the open source module keras.layers.LSTM in keras; The LSTM model parameters are optimized by minimizing the embedded prediction error, i.e. The working model based on the aging test data set is generated on the deep learning workstation by using the python programming according to the above method, and the hierarchical data format file HDF5 is output after the training is completed. (4) Convert the model generated on the deep learning workstation into a model based on C / C++ file, and compile and link the abnormality detection model into the aging detection control system; (5) Deploy the aging test abnormality detection model to the industrial computer to run, collect time series based current data in real time, convert through VAE encoder, and predict the VAE hidden variable corresponding to the next window through LSTM; represents the kth embedded output predicted by the LSTM model for the time series window; denotes the reconstructed time series; Decoder represents the VAE decoder; For the window of reconstruction, define a function R t that evaluates the test for abnormality by computing the prediction error w t ; w t-(k-i)*p denotes the original input time series; denotes the reconstructed time series; R t denotes the prediction error; Set a threshold, and calculate according to the above formula (4). If the result is greater than the set threshold, it is determined that aging abnormality occurs. After the aging test time window ends, a predictive alarm is made.

3. The mass production electric control PCBA aging test anomaly detection method of claim 1, wherein: The aging test box (31) comprises a test base plate (311) and a base plate connecting card slot (313). The PCBA (312) to be tested is connected with the test base plate (311) through the base plate connecting card slot (313). The test base plate (311) supplies power for the PCBA (312) to be tested.

4. The mass production electric control PCBA aging test anomaly detection method of claim 1, wherein: The industrial computer (26) is connected with the program-controlled power supply (21), the relay passage switch unit (22), the ammeter (23), the temperature control unit (24) and the humidity control unit (25) through the signal control bus (27) respectively, the program-controlled power supply (21) is connected with the relay passage switch unit (22) through the power supply power line (28), the power supply output is controlled through the relay passage switch unit (22), the relay passage switch unit (22) is connected with the ammeter (23) through the power supply power line (28), the current data is acquired in real time, the aging test box (31) is connected through the power supply power line (28), the current data of the power supply is acquired in real time through switching the current voltage output.

5. The mass production electric control PCBA aging test anomaly detection method of claim 1, wherein: The temperature control unit (24) of the industrial control unit (2) is connected with the heating electric heater and the temperature sensor group in the aging test room through the signal control bus (27), the humidity control unit (25) is connected with the humidifier and the humidity sensor group in the aging test room through the signal control bus (27), the industrial control unit is connected with the aging test box in the aging test room through the power supply power line (28).

6. The mass production electric control PCBA aging test anomaly detection method according to claim 1 or 4 or 5, characterized in that: The signal control bus (27) is the modbus bus based on RS485 or the deviceNet bus based on network.

7. The mass production electric control PCBA aging test anomaly detection method according to claim 4 or 5, characterized in that: The power supply power line (28) is a double-core shielded power line.

Images