Summary of the invention
For the deficiency for solving existing laser soldering quality detection technology, the present invention provides a kind of swashing based on temperature curve
Light solder quality determining method records the input power information of open loop control algorithm, obtains solder joint temperature during laser output
Degree evidence reflects welding process using temperature data, to realize that quality of welding spot detects automatically, reduces cost, improves efficiency.
The present invention adopts the following technical scheme that realization: a kind of laser soldering quality determining method based on temperature curve,
The following steps are included:
A, the input parameter for setting the open loop control algorithm of laser soldering machine, focuses to laser head and adjusts pad
Defocusing amount;
B, it is welded according to open loop control algorithm, records the input parameter of open loop control algorithm, acquisition opened loop control is calculated
The power features of method are as first part's feature;
C, the temperature data in solder soldering processes is acquired, temperature data is filtered, filtered temperature is extracted
It spends data characteristics and forms feature vector, as second part feature;
D, first part's feature is combined to form feature vector with second part feature, then feature vector is sent into and is classified
Device is judged, the testing result of solder joint welding quality is obtained.
Preferably, the power features of open loop control algorithm described in step b include three preheating section, welding section and soaking zone ranks
The input power of section and duration.
Preferably, step c the following steps are included:
During pad solder, temperature data in solder soldering processes is acquired, is filtered using Butterworth filter
Wave, the temperature curve after being filtered;
Feature extraction is carried out in terms of time domain and frequency domain two to the temperature curve after filtering processing, obtains second part spy
Sign.
Preferably, to the temperature curve after filtering processing, the feature extracted from time domain includes the highest that preheating section reaches
Temperature and the wave crest number for maximum temperature to welding end occur.
Compared with prior art, the invention has the following advantages:
(1) the input power information of control algolithm is added in quality testing feature extraction, is adapted to the input of opened loop control
Variation;
(2) directly using the temperature data of laser soldering machine control output, design additional system is avoided to obtain weldering
Indirect signal in termination process, using temperature data reflect welding process, to realize that quality of welding spot detects automatically, reduce manually at
This, improves detection efficiency;
(3) testing result can continue Optimized model, improve detection accuracy.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and illustrated embodiment is served only for explaining the present invention,
It is not intended to limit the scope of the present invention.
The welding of laser soldering machine, which is broadly divided into, is loaded into pad, setting control input parameter and quality testing three parts.This
Invention laser soldering quality determining method is based on temperature curve and is put under laser soldering machine after the complete tin cream of PCB pad point
Side, setting are welded after controlling the power inputted.Infrared temperature probe detects the temperature change in welding process, and acquires
Temperature data is used for training pattern;Entire welding process uses open loop control algorithm, and the invariable power of input includes three sections, that is, is divided into
Preheating section, welding section and soaking zone power inhibit preheating section scaling powder by the temperature data of Butterworth filtering processing acquisition
The influence volatilized to temperature data;Temperature data carries out feature extraction after pretreatment, and the feature set of extraction includes first
Dtex is sought peace second part feature.In order to introduce the difference of open loop control algorithm input parameter, the control of first part's collection apparatus
The power input information of algorithm, second part feature are the time domain and frequency domain information of temperature curve, are mentioned using wavelet-decomposing method
Take temperature energy spectrum as frequency domain information;Finally, building convolutional neural networks, convolutional neural networks are sent into the feature set of extraction
It is trained, obtains model;Model is used for the quality testing of subsequent pad, finally obtaining detection accuracy is 85%, and
It being capable of continuous iteration optimization model.
In the present embodiment, the laser soldering quality determining method based on temperature curve is as shown in figs. 1-7, comprising:
S1, it is loaded into pad: the pad for having put tin cream being put into the lower section card slot of laser soldering machine, is carried out using camera
Positioning completes pad and is registrated work.
S2, the control parameter for setting laser soldering machine carry out the focusing of laser head using black and white camera and adjust pad
Defocusing amount.
Setting control parameter is mainly to set the input parameter of open loop control algorithm, comprising: preheating section, welding section and heat preservation
The input power of section three phases and duration;In the present embodiment, three phases input power such as Fig. 2 of open loop control algorithm
It is shown.
Wherein, defocusing amount be distance of the laser head apart from solder joint, this step focus while adjust pad defocusing amount from
And quality determining method of the present invention is made to can adapt to the variation that open loop control algorithm inputs parameter.It is focused using black and white camera,
Data consumption is less, at low cost.
S3, it after setting the input parameter of open loop control algorithm, is welded according to open loop control algorithm, records open loop
The input parameter of control algolithm acquires the power features of open loop control algorithm as first part's feature.
Power features include preheating section, the input power of welding section and soaking zone three phases and duration, so that matter
Amount detection algorithm can adapt to the variation of open loop control algorithm input.
S4, the temperature data in solder soldering processes is acquired by infrared temperature sensor, temperature data is filtered
Processing extracts filtered temperature data feature and forms feature vector, as second part feature.
Since warm-up phase tin cream can melt, scaling powder volatilizees and generates smog, therefore infrared temperature sensor is by shadow
It rings, the temperature data in solder soldering processes collected is not the actual temperature of solder joint, and is shaken acutely, after being unfavorable for
Continuous analysis.The present invention is filtered infrared temperature sensor temperature data collected using Butterworth filter, and adjusts
Whole filter parameter inhibits warm-up phase because influencing brought by scaling powder volatilization well.
In the specific implementation process, this step mainly utilizes Butterworth filter to improve the volatilization pair of preheating section scaling powder
The influence of infrared temperature sensor acquisition data;Feature, frequency domain are extracted over the frequency domain to filtered temperature data
On using wavelet decomposition come the energy spectrum of Extracting temperature data, to obtain feature vector.This step can be divided into following two step
To realize:
S41, during pad solder, utilize infrared temperature sensor acquisition solder soldering processes in temperature data.It adopts
It is filtered with Butterworth filter, the temperature curve after being filtered.Specifically: pad is sent into laser soldering machine
In, infrared temperature sensor acquires the temperature data Data of solder soldering processes, and temperature data is the control output of welding process,
It is direct data, can be very good reflection welding process.
The mathematical model of Butterworth filter are as follows:
H (w) indicates the amplitude of filter, WsIt is stopband cutoff frequency, wpFor passband marginal frequency, n is the rank of filter
Number.
In filter design function buttord (), parameter wp, WsThe respectively passband of digital filter, stopband cutoff frequency
The normalized value of rate, wherein 0≤Wp≤ 1,0≤Ws≤ 1, Wp<WsIt is then designed as bandpass filter, in circuit, stopband frequency
Rate formula:According to sample frequency fsFeature, incorporating parametric requirement, continuously attempts to filter Wp、WsParameter value is looked into
See filter effect, it is final to determine the filtering parameter met the requirements.
In the present embodiment, the design parameter of filter is designed are as follows: sample frequency fsFor 1000HZ, cut-off frequecy of passband WpForStopband cutoff frequency WsForPass band damping RpFor 2dB, stopband attenuation RsFor 40dB.
After this step obtains temperature data Data, temperature data Data is filtered by Butterworth filter, is obtained
To filtered data Filter_data, to the filtered result of temperature curve as shown in figure 3, from the figure 3, it may be seen that Butterworth is filtered
Wave device effectively improves preheating section because scaling powder volatilization bring influences.That is, being filtered by Butterworth filter
The temperature curve of processing not only improves preheating section because of influence brought by scaling powder volatilization, and temperature curve is also more smooth.
S42, feature is extracted to the temperature curve after filtering processing, mainly carries out feature in terms of time domain and frequency domain two and mentions
It takes, obtains second part feature.
Since laser soldering system is different in the input power of the pad of different size, lead to maximum temperature and each rank
The parameters such as section duration are different, and temporal signatures can not reflect all information of welding process.Therefore, it is necessary to Extracting temperature curves
Frequency domain information, the present embodiment using wavelet decomposition obtain temperature curve different frequency energy spectrum.
Specifically: feature extraction is carried out to the Filter_data after filtering processing, in conjunction with the specific of laser soldering
Operating condition and weld characteristics, feature are extracted in terms of time domain and frequency domain two.The feature extracted in time domain has maximum temperature, and (reaction is pre-
The maximum temperature that hot arc reaches, excessively high ball easy to form) and occur maximum temperature to welding terminate wave crest number (reflection it is subsequent
The jitter conditions of welding);Wavelet decomposition is used on frequency domain, in the present embodiment, using " sym8 " small echo to the temperature after filtering processing
Line of writing music carries out two-layer decomposition, solves each Scale energy.
The mathematical model of wavelet decomposition are as follows:
Wherein, a is the zooming parameter of wavelet decomposition, and b is translation parameters, functionIt is scaling function, function ψ (x) is
Wavelet function.EjIndicate the energy on scale j, xi,jIndicate signal by after wavelet decomposition on scale j detail signal i-th
A numerical value.The decomposition result of wavelet decomposition is as shown in Figure 4.
S5, first part's feature is combined to form feature vector Vector with second part feature, then by feature vector
Vector is sent into classifier and is judged, obtains the testing result of solder joint welding quality.
The concrete mathematical model of classifier such as following formula:
Wherein, wijIt is feature weight, Θ is activation primitive threshold value, and f is activation primitive.To the feature vector of sample extraction,
It is multiplied with feature weight, the output result of activation primitive is compared with expected result.It is obtained by BP algorithm to optimize weight matrix
To the model for being suitable for present laser solder quality testing.
In the present embodiment, classifier as shown in figure 5, using include input layer, hidden layer and output layer totally 3 layers BP nerve
Network, the number of plies of neural network hidden layer are set as 10 layers.
Temperature data in subsequent welding process is obtained feature vector and be sent into training mistake using identical processing method
The stable classifier mathematical model of difference, classifies, obtains quality evaluation.
In the present embodiment, the error iteration situation of BP neural network in the training process is as shown in fig. 6, final error is restrained
Near 0.12;The test set of BP neural network in varied situations improve quality detection accuracy as shown in fig. 7, main distribution
Near 0.85;By test, for welding, effectively whether detection accuracy is 85% to the present invention.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.