A kind of method that designs single crystal super alloy solid solution system
Technical field
The present invention relates to design the nonlinear method of single crystal super alloy solid solution system, particularly a kind of method of the design single crystal super alloy solid solution system based on artificial neural network.
Background technology
Single crystal super alloy is owing to having higher warm ability, outstanding military service performance and the good antioxidant property of holding, and is at present and the preferred material that advanced engine holds warm turbine blade in the future.Because the requirement to use temperature is more and more higher, refractory element is as more and more large in the alloying consumption of W, Mo, Re, Ru etc., and in first-generation single crystal super alloy, refractory element total amount is about 14w.t.%, brings up to about 20w.t.% to the third generation during as CMSX-10.Meanwhile, the use of the second-phase forming element such as Al, Ti, Ta, Nb also approaches solid solubility limit.Also brought thus following shortcoming: solid solubility temperature is too high, low melting point phase initial melting temperature is too low, and solid solution window is narrow, and solid solution process is slow, and solid solution system design difficulty even can not realize solid solution.For example, the standard solid solution system of CMSX-10 is from 1315 DEG C of ladder-elevating temperatures to 1365 DEG C, and 50 DEG C of temperature spans, reach more than 40 hour when total.As can be seen here, realize and judge that fast can superalloy solid solution and Exact Design solid solution system, design and industrial production important in inhibiting for superalloy.
And traditional high temperature alloy research method, generally to pass through on alloy designs, mother alloy melting, single crystal casting, first fusing point metallographic, solid solubility temperature roll off the production line, solid solution system design and adjustment, just can obtain solid solution system.An entire flow generally needs the time even more of a specified duration half a year.Simultaneously conventional alloys method of design inevitably occurs that material can not realize the situation of solid solution, causes unnecessary loss.
Meanwhile, some linear systems that are used for simulating heterogeneity superalloy performance are developed, but along with alloy element range extension, Alloying Amount approaches superalloy design limit, and its composition and performance more show nonlinear feature.Major cause is that single crystal super alloy is γ-γ ' biphasic system, all alloy elements influence each other in two-phase, each element distribution constant is all changed, cause linear system interalloy element prematrix to occur obviously to change and increase simulation error.So for the wide variation of composition, nonlinear system simulation result closing to reality more, it is more accurate to judge.In non-linear system, Back-Propagation artificial neural network technology is the most ripe, and compatible and error shows more outstanding.
Summary of the invention
Can the object of the invention is to realize quick judgement superalloy solid solution and Exact Design solid solution system, proposes a kind of method of the design single crystal super alloy solid solution system based on artificial neural network.
The invention provides a kind of novel method of the single crystal super alloy solid solution system design based on artificial neural network, first by the characteristic temperature database of heterogeneity single crystal super alloy, training Back-Propagation artificial neural network, afterwards, the artificial neural network of having trained is inputted to alloying constituent to be measured, calculate characteristic temperature and design solid solution system.
Here the characteristic temperature of selecting is solidus temperature, liquidus temperature, freezing range, second-phase solvent temperature, low melting point phase initial melting temperature and heat treatment state initial melting temperature.Wherein, solidus temperature is the temperature that starts fusing on single crystal alloy equilibrium phase diagram; Liquidus temperature is the temperature melting completely on single crystal alloy equilibrium phase diagram; When freezing range is single crystal casting, dendrite form nuclear temperature is to the temperature head of interdendritic complete solidification temperature; Second-phase solvent temperature is the solvent temperature of coherence γ ' in single crystal super alloy; Low melting point phase initial melting temperature is the fusing point that the low melting point phase forming latter stage is solidified in interdendritic; Heat treatment state initial melting temperature is the initial melting temperature of alloy after elimination as cast condition microsegregation.
These solid solution system design method concrete steps are as follows:
Step 1, set up the database of the single crystal super alloy characteristic temperature of following composition range (w.t.%):
W |
0~8% |
Mo |
0~16% |
Ta |
0~9% |
Al |
4~8.5% |
Ti |
0~4% |
Nb |
0~2% |
Re |
0~7% |
Ru |
0~3% |
Co |
0~10% |
Cr |
0~10% |
Y |
0~0.1% |
Ni |
bal. |
Wherein, " bal. " is the weight percentage of Ni element, and its numerical value is 100% to deduct other element wt per-cent sums.
Select artificial neural network to be input as single crystal super alloy composition, be output as solidus temperature, liquidus temperature, freezing range, second-phase solvent temperature, low melting point phase initial melting temperature and heat treatment state initial melting temperature.Setting up artificial neural network is Back-Propagation neural network, adopts three-decker, comprises input layer, hidden layer and output layer; Input layer number is 12, and output layer neuron number is 6,14 of hidden layer neuron numbers, and input layer and output layer neurone use purelin function representation, and hidden layer uses log-sigmoid function representation.
Step 2, from database, select more than 300 data sample as learning sample, training of human artificial neural networks use error back-propagation algorithm, netinit condition is that (0,0.5) random assignment is to weight matrix, (0,1) random assignment is to threshold values matrix.Definition error e is:
Wherein, t
kfor the value of learning sample, y
kfor the output of artificial neural network output layer, k is number of training.
Design requirements is that training objective error is e<0.001.In the time that output error meets the target error of setting requirement, training stops, and preserves weight matrix and threshold values matrix.
Step 3: utilize the artificial neural network of having trained in step 2, input alloying constituent to be measured, calculate the characteristic temperature of this alloy to be measured, and design superalloy solid solution system.Concrete grammar is as follows:
Situation one: in the time of second-phase solvent temperature > heat treatment state initial melting temperature, single crystal super alloy can not solid solution;
Situation two: in the time of second-phase solvent temperature≤heat treatment state initial melting temperature and low melting point phase initial melting temperature > second-phase solvent temperature and freezing range >25 DEG C, single crystal super alloy solid solution system is: second-phase solvent temperature/16h;
Situation three: in the time of second-phase solvent temperature≤heat treatment state initial melting temperature and low melting point phase initial melting temperature > second-phase solvent temperature and freezing range≤25 DEG C, single crystal super alloy solid solution system is: second-phase solvent temperature/10h;
Situation four: in the time of second-phase solvent temperature≤heat treatment state initial melting temperature and low melting point phase initial melting temperature≤second-phase solvent temperature and freezing range >25 DEG C, single crystal super alloy solid solution system is: low melting point phase initial melting temperature/4h+ second-phase solvent temperature/16h;
Situation five: in the time of second-phase solvent temperature≤heat treatment state initial melting temperature and low melting point phase initial melting temperature≤second-phase solvent temperature and freezing range≤25 DEG C, single crystal super alloy solid solution system is: low melting point phase initial melting temperature/4h+ second-phase solvent temperature/10h;
The invention has the beneficial effects as follows:
The artificial neural network input terminus of the present invention's training has been contained most alloying constituent and content range that superalloy adopts, can realize quick judgement single crystal super alloy solid solution and Exact Design solid solution system, can be used for instructing single crystal super alloy design and industrial production.
Brief description of the drawings
Fig. 1 is solid solution system design method flow diagram provided by the invention;
Fig. 2 is the scatter diagram that in embodiment, systematic error changes with the variation of hidden layer neuron number, and an error hour hidden layer neuron number is 14;
Fig. 3 a is the as-cast structure pattern of single crystal super alloy described in embodiment 1;
Fig. 3 b is the microstructure morphology of the heat treatment state of single crystal super alloy described in embodiment 1;
Fig. 3 c is the as-cast structure pattern of single crystal super alloy in embodiment 2;
Fig. 3 d is the heat treatment state microstructure morphology of single crystal super alloy in embodiment 2;
Fig. 3 e is the as-cast structure pattern of single crystal super alloy in embodiment 3;
Fig. 3 f is the heat treatment state microstructure morphology of single crystal super alloy in embodiment 3.
Embodiment
Below in conjunction with drawings and Examples, the present invention is further described.
embodiment 1
Adopt method of design provided by the invention to carry out single crystal super alloy solid solution system design, concrete steps are as follows:
Step 1, set up the database of the single crystal super alloy characteristic temperature of following composition range (w.t.%):
W |
0~8% |
Mo |
0~16% |
Ta |
0~9% |
Al |
4~8.5% |
Ti |
0~4% |
Nb |
0~2% |
Re |
0~7% |
Ru |
0~3% |
Co |
0~10% |
Cr |
0~10% |
Y |
0~0.1% |
Ni |
bal. |
Select artificial neural network to be input as single crystal super alloy composition, be output as solidus temperature, liquidus temperature, freezing range, second-phase solvent temperature, low melting point phase initial melting temperature and heat treatment state initial melting temperature.Setting up artificial neural network is Back-Propagation neural network, adopts three-decker, comprises input layer, hidden layer and output layer; Input layer number is 12, and output layer neuron number is 6,14 of hidden layer neuron numbers.Input layer and output layer neurone use purelin function representation, and hidden layer uses log-sigmoid function representation.
This hidden layer neuron number selection principle is: systematic error minimizes.Systematic error is the error amount that artificial neural network frequency of training is greater than 1000 back balance systems.As shown in Figure 2, the artificial neural network obtaining for single crystal super alloy characteristic temperature database training, when hidden layer neuron number is 14() time, systematic error minimum, therefore selecting hidden layer neuron number is 14.
Step 2, from database, select 311 data sample training artificial neural networks, training of human artificial neural networks use error back-propagation algorithm, netinit condition is (0,0.5) random assignment is to weight matrix, (0,1) random assignment is to threshold values matrix, and training objective error is e<0.001.In the time that the satisfied setting of output error requires, training stops, and preserves weight matrix and threshold values matrix.
Step 3: utilize the artificial neural network of having trained in step 2, input following alloying constituent to be measured:
W |
0% |
Mo |
11% |
Ta |
6% |
Al |
7.6% |
Ti |
0% |
Nb |
0% |
Re |
3% |
Ru |
0% |
Co |
5% |
Cr |
0% |
Y |
0% |
Ni |
67.4% |
The characteristic temperature that calculates this composition is 1360 DEG C of 1389 DEG C of solidus temperatures, 1411.2 DEG C of liquidus temperatures, 22.2 DEG C of freezing ranges, 1371.2 DEG C of second-phase solvent temperatures, 1351 DEG C of low melting point phase initial melting temperatures and heat treatment state initial melting temperatures.Can obtain, 1360 DEG C of 1371.2 DEG C of > heat treatment state initial melting temperatures of second-phase solvent temperature, meet situation one, and alloy cannot be realized solid solution.As shown in Figure 3, as-cast structure is shown in Fig. 3 a to experimental result, and 1360 DEG C/50 hours microtextures of thermal treatment are shown in Fig. 3 b, not solid solution.Experimental result conforms to design solid solution system.
embodiment 2
Step 1, set up the database of the single crystal super alloy characteristic temperature of following composition range (w.t.%):
W |
0~8% |
Mo |
0~16% |
Ta |
0~9% |
Al |
4~8.5% |
Ti |
0~4% |
Nb |
0~2% |
Re |
0~7% |
Ru |
0~3% |
Co |
0~10% |
Cr |
0~10% |
Y |
0~0.1% |
Ni |
bal. |
Select artificial nerve network model to be input as single crystal super alloy composition, be output as solidus temperature, liquidus temperature, freezing range, second-phase solvent temperature, low melting point phase initial melting temperature and heat treatment state initial melting temperature.Setting up artificial neural network is Back-Propagation neural network, adopts three-decker, comprises input layer, hidden layer and output layer; Input layer number is 12, and output layer neuron number is 6,14 of hidden layer neuron numbers, and input layer and output layer neurone use purelin function representation, and hidden layer uses log-sigmoid function representation.
Step 2, from database, select 311 data samples, training of human artificial neural networks use error back-propagation algorithm, netinit condition is (0,0.5) random assignment is to weight matrix, (0,1) random assignment is to threshold values matrix, and training objective error is e<0.001.In the time that the satisfied setting of output error requires, training stops, and preserves weight matrix and threshold values matrix.
Step 3: utilize the artificial neural network of having trained in step 2, input following alloying constituent to be measured:
W |
5% |
Mo |
2% |
Ta |
6% |
Al |
7.6% |
Ti |
1% |
Nb |
0% |
Re |
1.5% |
Ru |
0% |
Co |
0% |
Cr |
0% |
Y |
0% |
Ni |
76.9% |
The characteristic temperature that calculates this alloy to be measured is 1345 DEG C of 1355.3 DEG C of solidus temperatures, 1391.2 DEG C of liquidus temperatures, 35 DEG C of freezing ranges, 1336.6 DEG C of second-phase solvent temperatures, 1333 DEG C of low melting point phase initial melting temperatures and heat treatment state initial melting temperatures.There is following relation:
1345 DEG C of 1336.6 DEG C≤heat treatment state of second-phase solvent temperature initial melting temperatures;
1336.6 DEG C of 1333 DEG C≤second-phase of low melting point phase initial melting temperature solvent temperatures;
Freezing range >25 DEG C
Meet situation four, single crystal super alloy solid solution system to be measured is: 1333 DEG C/4h+1336.6/16h.As-cast structure is shown in Fig. 3 c, and 1333 DEG C/4h+1336.6/16h thermal treatment microtexture is shown in Fig. 3 d, realizes solid solution.Experimental result conforms to the design result of solid solution system of the present invention.
embodiment 3
The difference of the present embodiment and embodiment 1,2 only exists: step 3: utilize the artificial neural network of having trained in step 2, input following target component:
W |
0% |
Mo |
9.5% |
Ta |
3% |
Al |
7.8% |
Ti |
0% |
Nb |
0% |
Re |
1.5% |
Ru |
0% |
Co |
0% |
Cr |
1.5% |
Y |
0.05% |
Ni |
76.65% |
The characteristic temperature that calculates this composition is 1352 DEG C of 1371.7 DEG C of solidus temperatures, 1401.8 DEG C of liquidus temperatures, 30 DEG C of freezing ranges, 1326.1 DEG C of second-phase solvent temperatures, 1330 DEG C of low melting point phase initial melting temperatures and heat treatment state initial melting temperatures.There is following relation:
1352 DEG C of 1326.1 DEG C≤heat treatment state of second-phase solvent temperature initial melting temperatures;
1326.1 DEG C of 1330 DEG C of > second-phase solvent temperatures of low melting point phase initial melting temperature;
Freezing range >25 DEG C,
Meet situation two, superalloy solid solution system is: 1326.1 DEG C/16h.As-cast structure is shown in Fig. 3 e, and 1327 DEG C/16h thermal treatment microtexture is shown in Fig. 3 f, realizes solid solution.Experimental result conforms to the design result of solid solution system of the present invention.
The A relating in specification sheets DEG C/Bh is illustrated at A temperature and is incubated B hour.
Single crystal super alloy characteristic temperature for training of human artificial neural networks derives from published or certified single crystal super alloy characteristic temperature data, and the predictable single crystal super alloy composition range of artificial neural network is the published or certified high temperature alloy composition scope of typing learning sample.For example, Re element published or confirmed that superalloy addition (w.t.%) is 0(trade mark PWA1480), 0(trade mark SRR99), 0(trade mark CMSX-6), 1.5(IC27 confirms), 1.5(IC21 confirms), 3(trade mark PWA1484), 3(trade mark ReneN5), 3(trade mark CMSX-4), 4(trade mark MC-NG), 5(trade mark ReneN6), 5(trade mark TMS-75), 5(trade mark TMS-162), 7(trade mark CMSX-10) etc., the predictable Re elemental range of artificial neural network is 0~7(w.t.%).In like manner obtain the predictable scope of other elements, this predictable range has contained most alloying constituent and content range that superalloy adopts.