Merge pull request #5 from PML-UCF/develop

Release version 0.0.2

pinn/__init__.py

@@ -46,3 +46,4 @@
 """
 
 from . import layers
+__version__ = '0.0.2'

pinn/layers/__init__.py

@@ -45,13 +45,22 @@
 """ PINN layers
 """
 
+from .util import interpolate
+
 from .core import getScalingDenseLayer
+from .core import inputsSelection
+
+from .core import TableInterpolation
 
 from .physics import StressIntensityRange
 from .physics import ParisLaw
+from .physics import SNCurve
+
+from .physics import WalkerModel
 
 from .rnn import CumulativeDamageCell
 
 del physics
 del rnn
 del core
+del util

pinn/layers/core.py

@@ -41,27 +41,112 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 # SOFTWARE.
 # ==============================================================================
-
 """ Core PINN layers
 """
+from tensorflow.python.keras.layers import Dense
+
+from tensorflow.python.keras import initializers
+from tensorflow.python.keras import regularizers
+from tensorflow.python.keras import constraints
 
-from tensorflow.keras.layers import Dense
+from tensorflow.python.keras.engine.base_layer import Layer
+
+from tensorflow.python.framework import tensor_shape
+from tensorflow.python.framework import common_shapes
 from tensorflow.python.framework import ops
 
-from tensorflow.linalg import diag as tfDiag
-from tensorflow.math import reciprocal
+from tensorflow.python.ops import array_ops
+
+from pinn.layers import interpolate
 
+from tensorflow import shape, expand_dims, constant, cast
 
-def getScalingDenseLayer(input_location, input_scale, dtype):
-    input_location    = ops.convert_to_tensor(input_location, dtype=dtype)
-    input_scale       = ops.convert_to_tensor(input_scale, dtype=dtype)
-    recip_input_scale = reciprocal(input_scale)
+import numpy as np
+
+
+def getScalingDenseLayer(input_location, input_scale):
+    recip_input_scale = np.reciprocal(input_scale)
 
-    waux = tfDiag(recip_input_scale)
+    waux = np.diag(recip_input_scale)
     baux = -input_location*recip_input_scale
 
-    dL = Dense(input_location.get_shape()[0], activation = None, input_shape = input_location.shape)
+    dL = Dense(input_location.shape[0], activation = None, input_shape = input_location.shape)
     dL.build(input_shape = input_location.shape)
     dL.set_weights([waux, baux])
     dL.trainable = False
     return dL
+
+
+def inputsSelection(inputs_shape, ndex):
+    if not hasattr(ndex,'index'):
+        ndex = list(ndex)
+    input_mask = np.zeros([inputs_shape[-1], len(ndex)])
+    for i in range(inputs_shape[-1]):
+        for v in ndex:
+            if i == v:
+                input_mask[i,ndex.index(v)] = 1
+
+    dL = Dense(len(ndex), activation = None, input_shape = inputs_shape, 
+               use_bias = False)
+    dL.build(input_shape = inputs_shape)
+    dL.set_weights([input_mask])
+    dL.trainable = False
+    return dL
+
+
+class TableInterpolation(Layer):
+    """ Table lookup and interpolation implementation.
+        Interrogates provided query points using provided table and outputs the interpolation result.
+        Remarks on this class:
+            - Only supports 2-D tables (f(x1,x2) = y)
+            - If a 1-D table is to be used, it needs to be converted to a 2-D table (see file /samples/core/table_lookup/run01_table_lookup_sample.py)
+            - Extrapolation is not supported (provide a table grid large enough for your case)
+            - Class returns limit values in case of extrapolation attempt.
+            - Provided tables should be equally spaced.
+    """
+    def __init__(self,
+                 kernel_initializer = 'glorot_uniform',
+                 kernel_regularizer=None,
+                 kernel_constraint=None,
+                 table_shape=(1,4,4,1),
+                 **kwargs):
+        if 'input_shape' not in kwargs and 'input_dim' in kwargs:
+            kwargs['input_shape'] = (kwargs.pop('input_dim'),)
+        super(TableInterpolation, self).__init__(**kwargs)
+        self.kernel_initializer = initializers.get(kernel_initializer)
+        self.kernel_regularizer = regularizers.get(kernel_regularizer)
+        self.kernel_constraint  = constraints.get(kernel_constraint)
+
+        self.table_shape = table_shape
+
+    def build(self, input_shape, **kwargs):
+        self.grid = self.add_weight("grid",
+                                      shape = self.table_shape,
+                                      initializer = self.kernel_initializer,
+                                      dtype = self.dtype,
+                                      trainable = self.trainable,
+                                      **kwargs)
+        self.bounds = self.add_weight("bounds",
+                                      shape = [2,2],
+                                      initializer = self.kernel_initializer,
+                                      dtype = self.dtype,
+                                      trainable = self.trainable,
+                                      **kwargs)
+        self.built = True
+
+    def call(self, inputs):
+        self.grid = ops.convert_to_tensor(self.grid, dtype=self.dtype)
+        self.bounds = ops.convert_to_tensor(self.bounds,dtype=self.dtype)
+        inputs = ops.convert_to_tensor(inputs, dtype=self.dtype)
+        queryPoints_ind = ((cast(shape(self.grid)[1:3], dtype=self.dtype))-constant(1.0))*(inputs-self.bounds[0])/(self.bounds[1]-self.bounds[0])
+        if common_shapes.rank(inputs) == 2:
+            queryPoints_ind = expand_dims(queryPoints_ind,0)
+        output = interpolate(self.grid, queryPoints_ind)
+        if common_shapes.rank(inputs) == 2:
+            output = array_ops.reshape(output,(array_ops.shape(output)[1],) + (array_ops.shape(output)[2],))
+        return output
+
+    def compute_output_shape(self, input_shape):
+        aux_shape = tensor_shape.TensorShape((None,1))
+
+        return aux_shape[:-1].concatenate(1)

pinn/layers/physics.py

@@ -49,12 +49,18 @@
 
 from tensorflow.python.keras.engine.base_layer import Layer
 
+#TODO: addept to tf2
+from tensorflow.compat.v1 import placeholder
+
+from tensorflow.python.ops import gen_math_ops, array_ops
+
+from tensorflow import reshape
+
 from tensorflow.python.keras import initializers
 from tensorflow.python.keras import regularizers
 from tensorflow.python.keras import constraints
 
 from tensorflow.python.framework import ops
-from tensorflow.python.ops import gen_math_ops
 from tensorflow.python.framework import tensor_shape
 from tensorflow.python.framework import common_shapes
 
@@ -83,9 +89,7 @@ def __init__(self,
 
     def build(self, input_shape):
         input_shape = tensor_shape.TensorShape(input_shape)
-        if input_shape[-1].value is None:
-            raise ValueError('The last dimension of the inputs to `StressIntensityRange` '
-                             'should be defined. Found `None`.')
+
         self.kernel = self.add_weight("kernel",
                                       shape = [1],
                                       initializer = self.kernel_initializer,
@@ -96,17 +100,21 @@ def build(self, input_shape):
         self.built = True
 
     def call(self, inputs):
-        inputs  = ops.convert_to_tensor(inputs, dtype=self.dtype)
-        if common_shapes.rank(inputs) is not 2:
+        inputs = ops.convert_to_tensor(inputs, dtype=self.dtype)
+        if common_shapes.rank(inputs) is not 2: 
             raise ValueError('`StressIntensityRange` only takes "rank 2" inputs.')
-        output = self.kernel*inputs[:,1]*gen_math_ops.sqrt(np.pi*inputs[:,0])
+
+        output = gen_math_ops.mul(self.kernel*inputs[:,1], gen_math_ops.sqrt(np.pi*inputs[:, 0]))
+        output = array_ops.reshape(output, (array_ops.shape(output)[0], 1))
+
+        # outputs should be (None, 1), so it is still rank = 2
         return output
 
     def compute_output_shape(self, input_shape):
         aux_shape = tensor_shape.TensorShape((None,1))
         return aux_shape[:-1].concatenate(1)
 
-		
+
 class ParisLaw(Layer):
     """Just your regular Paris law implementation.
     `ParisLaw` implements the operation:
@@ -145,4 +153,103 @@ def call(self, inputs):
     def compute_output_shape(self, input_shape):
         aux_shape = tensor_shape.TensorShape((None,1))
         return aux_shape[:-1].concatenate(1) 
-
+
+
+class SNCurve(Layer):
+    """ SN-Curve implementation (REF: https://en.wikipedia.org/wiki/Fatigue_(material)#Stress-cycle_(S-N)_curve)
+        `output = 1/10**(a*inputs+b)`
+        where:
+            * `a`,`b` parametric constants for linear curve,
+            * input is cyclic stress, load, or temperature (depends on the application) in log10 space,
+            * output is delta damage
+        Notes:
+            * This layer represents SN-Curve linearized in log10-log10 space
+            * (a*inputs+b) expression gives number of cycles in log10 space corresponding to stress level
+        Linearization:
+            * For an SN-Curve with an equation of N = C1*(S**C2) , take log10 of both sides
+            * log10(N) = log10(C1) + C2*log10(S), yields to:
+                C2 = a
+                log10(C1) = b
+                log10(S) = inputs            
+    """
+    def __init__(self,
+                 kernel_initializer = 'glorot_uniform',
+                 kernel_regularizer=None,
+                 kernel_constraint=None,
+                 **kwargs):
+        if 'input_shape' not in kwargs and 'input_dim' in kwargs:
+            kwargs['input_shape'] = (kwargs.pop('input_dim'),)
+        super(SNCurve, self).__init__(**kwargs)
+        self.kernel_initializer = initializers.get(kernel_initializer)
+        self.kernel_regularizer = regularizers.get(kernel_regularizer)
+        self.kernel_constraint  = constraints.get(kernel_constraint)
+
+    def build(self, input_shape, **kwargs):
+        self.kernel = self.add_weight("kernel",
+                                      shape = [2],
+                                      initializer = self.kernel_initializer,
+                                      dtype = self.dtype,
+                                      trainable = self.trainable,
+                                      **kwargs)
+        self.built = True
+
+    def call(self, inputs):
+        output = 1/10**(self.kernel[0]*inputs+self.kernel[1])
+        return output
+
+    def compute_output_shape(self, input_shape):
+        aux_shape = tensor_shape.TensorShape((None,1))
+        return aux_shape[:-1].concatenate(1) 
+
+
+class WalkerModel(Layer):
+    """A modified version of Paris law to take into account the stress ratio effect.
+    `WalkerModel` implements the operation:
+        `output = C*(inputs[:,0]**m)`
+        where `C` and `m` are constants, and `C` is obtained from the following
+        relation:
+            `C = Co/((1-inputs[:,1])**(m*(1-gamma))))`
+            
+            * input[:,0] is the nominal stress range
+            * input[:,1] is the stress ratio, and
+            
+            * sig is a custumized sigmoid function to calibrate Walker's coefficient (gamma)
+                  with respect to the stress ratio value.        
+    """
+    def __init__(self,
+                 kernel_initializer = 'glorot_uniform',
+                 kernel_regularizer=None,
+                 kernel_constraint=None,
+                 **kwargs):
+        if 'input_shape' not in kwargs and 'input_dim' in kwargs:
+            kwargs['input_shape'] = (kwargs.pop('input_dim'),)
+        super(WalkerModel, self).__init__(**kwargs)
+        self.kernel_initializer = initializers.get(kernel_initializer)
+        self.kernel_regularizer = regularizers.get(kernel_regularizer)
+        self.kernel_constraint  = constraints.get(kernel_constraint)
+
+    def build(self, input_shape, **kwargs):
+        self.kernel = self.add_weight("kernel",
+                                      shape = [4],
+                                      initializer = self.kernel_initializer,
+                                      dtype = self.dtype,
+                                      trainable = True,
+                                      **kwargs)
+        self.built = True
+
+    def call(self, inputs):
+        inputs = ops.convert_to_tensor(inputs, dtype=self.dtype)
+        if common_shapes.rank(inputs) is not 2: 
+            raise ValueError('`WalkerModel` only takes "rank 2" inputs.')
+
+        sig = 1/(1+gen_math_ops.exp(self.kernel[0]*inputs[:,1]))
+        gamma = sig*self.kernel[1]
+        C = self.kernel[2]/((1-inputs[:,1])**(self.kernel[3]*(1-gamma)))
+        output = C*(inputs[:,0]**self.kernel[3])
+        output = array_ops.reshape(output,(array_ops.shape(output)[0],1))
+        return output
+
+    def compute_output_shape(self, input_shape):
+        aux_shape = tensor_shape.TensorShape((None,1))
+        return aux_shape[:-1].concatenate(1)
+

pinn/layers/rnn.py

@@ -86,7 +86,7 @@ def get_initial_state(self, inputs=None, batch_size=None, dtype=None):
         if self.initial_damage is None:
             initial_state = _generate_zero_filled_state_for_cell(self, inputs, batch_size, dtype)
         else:
-            initial_state = self.initial_damage
+            initial_state = ops.convert_to_tensor(self.initial_damage, dtype=self.dtype)
 
         return initial_state