R caret: How do I apply separate pca to different dataframes before training?

I use caret in R. My final goal is to submit different dataframes to separate preProcess pca and then put the PCA-components together in one training with ridge regression. However, see example code below where I don't get the same results when applying pca in preProcess within versus outside/before train function.

1. Why do I not get the same results?
2. And how do I achieve my main goal in the best way?

#Sample data
s <- c(-0.412440717220306, -0.459911376237869, -0.234769582748413, -0.332282930612564, -0.486973077058792, -0.301480442285538, -0.181094691157341, -0.240918189287186, 0.0962697193026543, -0.119731709361076, -0.389783203601837, -0.217093095183372, -0.302948802709579, -0.406619131565094, 0.247409552335739, -0.406119048595428, 0.0574243739247322, -0.301231145858765, -0.229316398501396, -0.0620433799922466)
t <- c(0.20061232149601, 0.0536709427833557, 0.530373573303223, 0.523406386375427, 0.267315864562988, 0.413556098937988, 0.274257719516754,  0.275401413440704, 0.634453296661377, 0.145272701978683, 0.196711808443069, 0.332845687866211, 0.345706522464752, 0.444085538387299, 0.253269702196121, 0.231440827250481, -0.196317762136459, 0.49691703915596, 0.43754768371582, 0.0106721892952919)
u <- c(-0.565160751342773, 0.377725303173065,-0.273447960615158, -0.338064402341843, -0.59904420375824, -0.780133605003357,-0.508388638496399, -0.226167500019073, -0.257708549499512, -0.349863946437836,-0.443032741546631, -0.36387038230896, -0.455201774835587, -0.137616977095604,0.130770832300186, -0.420618057250977, -0.125859051942825, -0.382272869348526, -0.355217516422272, -0.0601325333118439)
v <- c(-0.45850995182991, -0.0105021595954895, -0.475157409906387, -0.325350821018219, -0.548444092273712, -0.562069535255432, -0.473256289958954, -0.492668628692627, -0.205974608659744, -0.266964733600616, -0.289298176765442, -0.615423858165741, -0.261823982000351, -0.472221553325653,  -0.684594392776489, -0.42777806520462, -0.240604877471924, -0.589631199836731, -0.782602787017822, -0.468854814767838)
w <- c(-0.886135756969452, -0.96577262878418,-0.755464434623718, -0.640497982501984, -0.849709093570709, -0.837802410125732, -0.659287571907043, -0.646972358226776, 0.0532735884189606, -0.646163880825043,-0.963890254497528, -0.91286826133728, -1.10484659671783, -0.596551716327667,  -0.371927708387375, -0.684276521205902, -0.55376398563385, -0.969008028507233, -0.956810772418976, -0.0229262933135033)

y <- c(9, 26, 30, 15, 25, 30, 30, 35, 35, 30, 21, 30,  9, 33, 31, 34, 29, 35, 25, 31)

#Sample data for procedure 1 and 2
df_test1 <- data.frame(s, t, u, v, w)
df_test2 <- df_test1

#PROCEDURE 1: preProcess (pca) applied WITHIN "train" function
library(caret)    
ytrain_df_test <- c(1:nrow(df_test1)) # number of observation that should be split in to the number of folds.
ntrain <- length(ytrain_df_test)    
# define folds
cv_folds <- createFolds(ytrain_df_test, k = 10, list = TRUE, returnTrain = TRUE) #, ...)
# define training control
train_control <- trainControl(method="cv", index = cv_folds, savePredictions = 'final') #, ...)
#adding y
df_test1$y <- y
# train the model
set.seed(1)
model1 <- caret::train(y~., data=df_test1, trControl=train_control, method= 'ridge', preProcess = 'pca')
output1 <- list(model1, model1$pred, summary(model1$pred), cor.test(model1$pred$pred, model1$pred$obs))
names(output1) <- c("Model", "Model_pred", "Summary", "Correlation")
output1


#PROCEDURE 2: preProcess (pca) applied OUTSIDE/BEFORE "train" function
ytrain_df_test <- c(1:nrow(df_test2)) # number of observation that should be split in to the number of folds.
ntrain <- length(ytrain_df_test)    

df2 <- preProcess(df_test2, method="pca", thresh = 0.95)
df_test2 <- predict(df2, df_test2)
df_test2$y <- y
df_test2
# define folds
cv_folds <- createFolds(ytrain_df_test, k = 10, list = TRUE, returnTrain = TRUE) 
# define training control
train_control <- trainControl(method="cv", index = cv_folds, savePredictions = 'final') 
# train the model
set.seed(1)
model2 <- caret::train(y~., data=df_test2, trControl=train_control, method= 'ridge') #, preProcess = 'pca')
model2
output2 <- list(model2, model2$pred, summary(model2$pred), cor.test(model2$pred$pred, model2$pred$obs))
names(output2) <- c("Model", "Model_pred", "Summary", "Correlation")
output2```