‭ I for ME‬

A
‭Department of Mechanical Engineering‬

‭Answers:‬

‭1.‬ C‭ ompare‬‭Depth‬‭First‬‭Search‬‭(DFS),‬‭Breadth‬‭First‬‭Search‬‭(BFS),‬‭and‬‭Hill‬‭Climbing‬
‭algorithms.‬‭Explain‬‭one‬‭scenario‬‭in‬‭mechanical‬‭engineering‬‭where‬‭each‬‭might‬‭be‬
‭beneficial.‬

‭Comparison of DFS, BFS, and Hill Climbing Algorithms‬

‭(Refer slides 191 & 192 in Module - 1 PPT)‬

‭Algorithm‬ ‭Key Characteristics‬ ‭Strengths‬ ‭Weaknesses‬

‭ epth‬ ‭First‬ E
D ‭ xplores‬ ‭as‬ ‭far‬ ‭as‬ ‭possible‬ ‭ fficient‬ ‭in‬ ‭memory‬
E ‭ an‬ ‭get‬ ‭stuck‬ ‭in‬
C
‭Search‬ ‭along‬ ‭each‬ ‭branch‬ ‭before‬ ‭usage;‬ ‭good‬ ‭for‬ ‭infinite‬ ‭loops‬ ‭in‬ ‭cyclic‬
‭(DFS)‬ ‭backtracking.‬‭Follows‬‭a‬‭LIFO‬ ‭problems‬ ‭where‬ ‭the‬ ‭graphs‬ ‭if‬ ‭not‬
‭(Last‬ ‭In,‬ ‭First‬ ‭Out)‬ ‭solution‬ ‭lies‬ ‭deep‬ ‭in‬ ‭implemented‬ ‭with‬
‭approach using a stack.‬ ‭the search tree.‬ ‭visited‬‭nodes‬‭or‬‭depth‬
‭limits.‬

‭ readth‬
B ‭ xplores‬ ‭all‬ ‭nodes‬ ‭at‬ ‭the‬
E ‭ uarantees‬
G ‭the‬ ‭ igh‬ ‭memory‬ ‭usage‬
H
‭First‬ ‭current‬ ‭depth‬ ‭before‬ ‭moving‬ ‭shortest‬ ‭path‬ ‭in‬ ‭due‬ ‭to‬ ‭storing‬ ‭all‬
‭Search‬ ‭to‬ ‭the‬ ‭next.‬ ‭Follows‬ ‭a‬ ‭FIFO‬ ‭unweighted‬ ‭graphs;‬ ‭nodes‬ ‭at‬ ‭a‬ ‭given‬
‭(BFS)‬ ‭(First‬ ‭In,‬ ‭First‬ ‭Out)‬ ‭systematically‬ ‭depth.‬
‭approach using a queue.‬ ‭examines all nodes.‬

‭ ill‬
H ‭ ‬ ‭heuristic‬ ‭search‬ ‭algorithm‬
A ‭ fficient‬ ‭for‬ f‭inding‬ ‭Can‬ ‭get‬ ‭stuck‬ ‭in‬‭local‬
E
‭Climbing‬ ‭that‬‭moves‬‭in‬‭the‬‭direction‬‭of‬ ‭solutions‬ ‭in‬ ‭search‬ ‭maxima,‬ ‭plateaus,‬ ‭or‬
‭increasing‬ ‭value‬ ‭(closer‬ ‭to‬ ‭spaces‬ ‭with‬ ‭a‬ ‭clear‬ r‭ idges;‬ ‭does‬ ‭not‬
‭the‬ ‭goal)‬ ‭until‬ ‭no‬ ‭further‬ ‭gradient.‬ ‭guarantee‬ ‭finding‬ ‭the‬
‭improvement is possible.‬ ‭optimal solution.‬

‭Applications in Mechanical Engineering‬

‭1.‬ D ‭ epth First Search (DFS)‬

‭Scenario:‬‭Stress Analysis of a Complex Mechanical‬‭Structure‬
‭When analyzing stress propagation through interconnected components of a mechanical‬
‭structure, DFS can traverse the structure to identify points of high stress concentration.‬
‭○‬ ‭Example:‬‭In a network of beams, DFS can explore each‬‭branch deeply to‬
‭determine how stress flows through the structure, identifying failure-prone points‬
‭in deep substructures.‬
‭2.‬ ‭Breadth First Search (BFS)‬
‭Scenario:‬‭Pathfinding in a Robotic Arm's Motion‬

‭6‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭ FS is ideal for determining the shortest path for a robotic arm to move from its initial‬
B
‭position to a target in a grid-based or discrete space.‬
‭○‬ ‭Example:‬‭For a welding robot working on an assembly line, BFS ensures the‬
‭arm moves efficiently without collisions, covering all welding points‬
‭systematically.‬
‭3.‬ ‭Hill Climbing‬
‭Scenario:‬‭Optimization of Cutting Tool Parameters‬‭in CNC Machining‬
‭Hill Climbing can be used to optimize tool speed, feed rate, and depth of cut to achieve‬
‭the best surface finish and minimize machining time.‬
‭○‬ ‭Example:‬‭By iteratively adjusting parameters and evaluating‬‭the quality of the‬
‭resulting cut, the algorithm can identify the most effective machining settings,‬
‭though it may require a workaround for avoiding local optima.‬

‭2.‬ E‭ xplain‬ ‭the‬ ‭key‬ ‭components‬ ‭of‬ ‭problem‬ ‭formulation‬ ‭in‬ ‭AI‬ ‭and‬ ‭apply‬ ‭them‬ ‭to‬ ‭a‬
‭real-world scenario such as Drone Delivery of Packages.‬

‭(Refer Slides 137 to 157 in Module - 1 PPT)‬

‭Refer Python Code uploaded in the Google Classroom.‬

‭Solution:‬

‭Key Components of Problem Formulation in AI‬

‭1.‬ I‭nitial State‬

‭Defines where the agent starts, including all relevant details about its‬
‭environment.‬
‭○‬ ‭For Drone Delivery: Initial location of the drone, battery level, and map of‬
‭no-fly zones.‬
‭2.‬ ‭Actions‬
‭Possible moves the agent can make to transition between states.‬
‭○‬ ‭For Drone Delivery: Taking off, landing, and moving in various directions‬
‭(e.g., up, down, left, right).‬
‭3.‬ ‭Goal State‬
‭Describes the desired outcome or the solution the agent must achieve.‬
‭○‬ ‭For Drone Delivery: Deliver the package to the specified customer‬
‭location and return to the base if necessary.‬
‭4.‬ ‭Path Cost‬
‭Assigns a cost to each action or transition to measure the efficiency of the path.‬
‭○‬ ‭For Drone Delivery: Cost could depend on battery usage, distance‬
‭covered, or time taken.‬
‭5.‬ ‭Constraints‬
‭Conditions the agent must adhere to while performing actions.‬

‭7‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭○‬ F ‭ or Drone Delivery: Avoid no-fly zones, ensure enough battery life to‬
‭complete the task, and adhere to air traffic regulations.‬
‭6.‬ ‭Criteria for Evaluating Solutions‬
‭Metrics to assess and compare solutions to ensure optimal or satisfactory‬
‭outcomes.‬
‭○‬ ‭For Drone Delivery: Minimize battery consumption, avoid obstacles, and‬
‭deliver within the shortest possible time.‬

‭Application to Drone Delivery of Packages‬

‭Scenario Setup‬

‭ ‬‭drone‬‭must‬‭deliver‬‭a‬‭package‬‭from‬‭a‬‭warehouse‬‭to‬‭a‬‭customer’s‬‭home.‬‭It‬‭starts‬‭with‬‭a‬
A
‭fully‬ ‭charged‬ ‭battery,‬ ‭and‬ ‭the‬ ‭map‬ ‭includes‬ ‭specific‬ ‭no-fly‬ ‭zones.‬ ‭The‬ ‭drone‬ ‭must‬
‭calculate‬ ‭the‬ ‭most‬ ‭efficient‬ ‭path‬ ‭to‬‭the‬‭customer‬‭while‬‭conserving‬‭energy‬‭and‬‭avoiding‬
‭restricted areas.‬

‭Key Components Applied‬

‭1.‬ ‭Initial State:‬

‭○‬ ‭Drone starts at the warehouse at coordinates (0, 0).‬
‭○‬ ‭Battery level: 100%.‬
‭○‬ ‭No-fly zones: [(2, 3), (3, 3)].‬
‭○‬ ‭Goal location: Customer's home at coordinates (5, 5).‬
‭2.‬ ‭Actions:‬
‭○‬ ‭Takeoff: Drone becomes airborne and starts the mission.‬
‭○‬ ‭Move: Drone moves in one of four directions (up, down, left, right).‬
‭○‬ ‭Land: Drone lands to deliver the package or return to the warehouse.‬
‭3.‬ ‭Goal State:‬
‭○‬ ‭Reach the goal location (5, 5) and successfully deliver the package.‬
‭4.‬ ‭Path Cost:‬
‭○‬ ‭Each movement reduces the battery by 1 unit.‬
‭○‬ ‭Additional penalties for detours or inefficient paths.‬
‭5.‬ ‭Constraints:‬
‭○‬ ‭Drone must avoid no-fly zones.‬
‭○‬ ‭Must not deplete the battery below 20% to ensure a safe return.‬
‭6.‬ ‭Criteria for Evaluating Solutions:‬
‭○‬ ‭The solution is optimal if the drone reaches the goal using the least‬
‭battery and shortest time while avoiding constraints.‬

‭Implementation Example‬

‭The drone executes the following steps:‬

‭●‬ ‭Takeoff: Drone transitions to airborne mode.‬

‭8‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭‬ M
● ‭ ove: Drone moves through the grid, avoiding no-fly zones.‬
‭●‬ ‭Deliver: Drone lands at the goal location.‬

‭Output Example:‬

‭‬ F
● ‭ inal Location: Customer’s Home (5, 5)‬
‭●‬ ‭Remaining Battery: 95%‬
‭●‬ ‭Objective Function Score: 95‬

‭ his‬ ‭problem‬ ‭formulation‬ ‭ensures‬ ‭the‬ ‭drone‬ ‭can‬ ‭make‬ ‭decisions‬ ‭autonomously‬ ‭while‬
T
‭adhering‬ ‭to‬ ‭constraints.‬ ‭Proactive‬ ‭planning‬ ‭by‬ ‭the‬ ‭agent,‬ ‭such‬ ‭as‬‭conserving‬‭batteries‬
‭for‬‭return‬‭trips‬‭and‬‭avoiding‬‭no-fly‬‭zones,‬‭highlights‬‭the‬‭importance‬‭of‬‭effective‬‭problem‬
‭formulation in achieving optimal solutions.‬

‭3.‬ E‭ xplain‬‭problem‬‭graphs‬‭and‬‭discuss‬‭its‬‭role‬‭in‬‭AI‬‭problem-solving.‬‭Illustrate‬‭with‬
‭an example of how it can be applied in a mechanical engineering context.‬

‭ roblem graphs are visual representations of possible states and actions for a given‬
P
‭problem.‬
‭They serve as a map of the problem, where nodes represent states (situations or‬
‭configurations of the problem), and edges (connections between nodes) represent‬
‭actions or transitions that move the system from one state to another.‬
‭●‬ ‭S – Start State‬
‭●‬ ‭G –Goal State‬
‭●‬ ‭Edges: Directed Arrows connecting each node‬
‭●‬ ‭A, B, C, D: Intermediate nodes or states.‬

‭Structure of a Problem Graph:‬

‭1.‬ N
‭ odes: Each node in a problem graph represents a possible state of the‬
‭problem.‬
‭a.‬ ‭In a puzzle, each node might represent a specific arrangement of pieces.‬
‭b.‬ ‭In a navigation problem, each node could represent a specific location.‬

‭9‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭2.‬ E ‭ dges: Each edge between nodes represents an action or transition from one‬
‭state to another.‬
‭a.‬ ‭The edge might include a cost (like distance, time, or resource use)‬
‭associated with moving from one state to the next.‬
‭3.‬ ‭Initial State: The starting point of the problem, often represented by a specific‬
‭node.‬
‭4.‬ ‭Goal State(s): The desired end state(s) we aim to reach, represented by one or‬
‭more nodes.‬
‭5.‬ ‭Path: A series of connected nodes from the initial state to a goal state,‬
‭representing a possible solution.‬

‭Example of a Problem Graph in Mechanical Engineering‬

‭Context: A robot navigating a manufacturing plant to deliver parts.‬

‭ ‬ I‭nitial State: The robot starts at its current location (e.g., Station A).‬
●
‭●‬ ‭Goal State: The robot needs to deliver parts to Station C.‬
‭●‬ ‭States (Nodes): Represent the robot's locations (e.g., Station A, Station B,‬
‭Station C).‬
‭●‬ ‭Actions (Edges): The paths the robot can take between stations.‬
‭●‬ ‭Path Costs: Represent time, energy, or distance required to move between‬
‭locations.‬

‭Illustration‬

‭Node (States)‬ ‭Actions (Edges)‬ ‭Cost (Path Weight)‬

‭Station A‬ ‭Move to Station B‬ ‭5 units of energy‬

‭Station B‬ ‭Move to Station C‬ ‭3 units of energy‬

‭Station A‬ ‭Move to Station C‬ ‭10 units of energy‬

‭In this graph:‬

‭ .‬ T
1 ‭ he robot starts at Station A (Initial State).‬
‭2.‬ ‭It can move to Station B or directly to Station C.‬
‭3.‬ ‭Path Costs are assigned to each edge (e.g., 5 units of energy to reach Station B‬
‭and 3 more to reach Station C).‬

‭4.‬ D‭ escribe‬ ‭the‬ ‭heuristic‬ ‭function‬ ‭in‬ ‭AI.‬ ‭Develop‬ ‭a‬ ‭heuristic‬ ‭function‬ ‭for‬ ‭a‬ ‭robot‬
‭navigating‬ ‭a‬ ‭grid‬ ‭maze‬ ‭where‬ ‭the‬ ‭objective‬ ‭is‬ ‭to‬ ‭minimize‬ ‭the‬ ‭number‬ ‭of‬ ‭steps‬
‭taken.‬

‭10‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭●‬ H ‭ euristics‬‭guide‬‭the‬‭AI‬‭by‬‭providing‬‭approximate‬‭solutions‬‭or‬‭shortcuts,‬‭making‬‭it‬
‭easier to define complex problems where exact solutions are impractical.‬
‭●‬ ‭Heuristics are often defined based on expert knowledge of the problem.‬
‭●‬ ‭Heuristic functions allow the AI to evaluate and prioritize paths or actions,‬
‭speeding up the search for solutions. Heuristics reduce the search space by‬
‭focusing on the most promising options first.‬
‭Key Characteristics of Heuristic Functions:‬
‭1.‬ ‭Informed Guess‬
‭2.‬ ‭Unlike brute-force search, a heuristic doesn’t explore all possibilities; it uses‬
‭additional information (the heuristic) to make better guesses about which paths to‬
‭explore first.‬
‭3.‬ ‭Estimation:‬
‭4.‬ ‭Heuristics don't guarantee an exact solution but offer a quick approximation that‬
‭helps reduce search time.‬
‭5.‬ ‭Problem-Specific:‬
‭6.‬ ‭Heuristic functions are often tailored to specific problems, leveraging knowledge‬
‭about the problem’s structure to make better estimates.‬

‭Heuristic Function for a Robot Navigating a Grid Maze‬

I‭magine‬ ‭a‬‭robot‬‭navigating‬‭a‬‭grid‬‭maze‬‭(like‬‭a‬‭chessboard)‬‭where‬‭it‬‭can‬‭only‬‭move‬‭up,‬
‭down, left, or right to reach a goal position.‬

‭The goal of the robot is to reach the destination (goal state) in the fewest possible steps.‬

‭How It Works:‬

‭ .‬ C
1 ‭ urrent Position: Where the robot is currently located.‬
‭2.‬ ‭Goal Position: The target where the robot needs to reach.‬
‭3.‬ ‭Heuristic Function: Estimates the number of moves (or steps) needed to reach‬
‭the goal from the current position.‬

‭Example - Robot on a Grid Maze‬

‭Consider a grid with rows and columns, like this:‬

‭0‬ ‭1‬ ‭2‬ ‭3‬

‭0‬ ‭Start‬

‭1‬

‭11‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭2‬

‭3‬ ‭Goal‬

‭‬ S
● ‭ tart Position: (0, 0) → Top-left corner‬
‭●‬ ‭Goal Position: (3, 3) → Bottom-right corner‬

‭Estimating the Steps‬

‭The simplest heuristic for this problem is the Manhattan Distance:‬

‭Heuristic = Steps to move horizontally + Steps to move vertically‬

‭●‬ A ‭ t the Start Position (0, 0):‬

‭To reach (3, 3), the robot needs:‬
‭○‬ ‭3 steps down (vertical), and‬
‭○‬ ‭3 steps right (horizontal).‬
‭●‬ ‭Total Steps = 3 + 3 = 6.‬
‭●‬ ‭If the robot moves to (1, 0):‬
‭○‬ ‭Steps to Goal: 2 down, 3 right → Total = 5 steps.‬
‭●‬ ‭If the robot moves to (2, 2):‬
‭○‬ ‭Steps to Goal: 1 down, 1 right → Total = 2 steps.‬
‭●‬ ‭At the Goal Position (3, 3):‬
‭○‬ ‭Steps to Goal = 0 (already there).‬

‭Role of Heuristic Function‬

‭●‬ A ‭ t each step, the robot calculates the estimated steps (heuristic value) to reach‬
‭the goal.‬
‭●‬ ‭The robot uses this estimate to decide which direction to move next:‬
‭○‬ ‭If moving right decreases the steps to the goal, the robot will move right.‬
‭○‬ ‭If moving down decreases the steps to the goal, the robot will move down.‬

‭5.‬ E
‭ xplain the role of convolutional neural networks (CNNs) in computer vision‬
‭tasks. Develop a conceptual architecture for a CNN model to detect defects in‬
‭manufactured parts.‬

‭●‬ C
‭ onvolutional neural networks provide a scalable approach to image‬
‭classification and object recognition tasks.‬

‭12‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭●‬ U ‭ ses principles from linear algebra, specifically matrix multiplication, to identify‬
‭patterns within an image.‬
‭●‬ ‭Uses high computational power‬
‭●‬ ‭Requiring graphical processing units (GPUs) to train models.‬
‭●‬ ‭Convolutional neural networks are distinguished from other neural networks by‬
‭their superior performance with image, speech, or audio signal inputs.‬
‭●‬ ‭They have three main types of layers, which are:‬
‭1.‬ ‭Convolutional layer‬
‭2.‬ ‭Pooling layer‬
‭3.‬ ‭Fully-connected (FC) layer‬

‭Conceptual Architecture of a CNN Model to Detect Defects in Manufactured Parts‬

‭ or‬ ‭detecting‬ ‭defects‬ ‭in‬ ‭manufactured‬ ‭parts‬ ‭(e.g.,‬ ‭cracks,‬ ‭missing‬ ‭components,‬ ‭or‬
F
‭scratches), a typical CNN model can follow this architecture:‬

‭1. Input Layer‬

‭ ‬ ‭Purpose: Takes in the image of the manufactured part.‬

●
‭●‬ ‭Input Size: For example, a grayscale image of size 128x128 pixels.‬

‭2. Convolutional Layer‬

‭●‬ P ‭ urpose: Applies filters (kernels) to the image to detect features like edges, lines,‬
‭or shapes.‬
‭●‬ ‭Operation: Performs a mathematical operation called convolution between the‬
‭image and the filter.‬
‭●‬ ‭Output: A feature map that highlights detected features.‬

‭13‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭Example: A 3x3 filter scans across the image to identify edges or small patterns.‬

‭3. Activation Layer (ReLU)‬

‭●‬ P ‭ urpose: Introduces non-linearity to the network, allowing it to learn complex‬

‭relationships.‬
‭●‬ ‭Operation: Applies the ReLU (Rectified Linear Unit) activation function to remove‬
‭negative values.‬

‭ReLU(x) = max(0,x)‬

‭4. Pooling Layer (Max Pooling)‬

‭●‬ P ‭ urpose‬‭: Reduces the size of the feature maps (downsampling) to make‬
‭computations faster and reduce overfitting.‬
‭●‬ ‭Operation‬‭: Takes the maximum value from a small region (e.g., 2x2) of the‬
‭feature map.‬

‭ xample: A 2x2 pooling window selects the largest value in each region, summarizing‬
E
‭the data.‬

‭5. Additional Convolutional and Pooling Layers‬

‭●‬ ‭Purpose‬‭: Extracts more complex features at deeper levels.‬

‭○‬ ‭Convolution Layers‬‭: Learn higher-level features like cracks, missing‬
‭teeth, or irregular shapes.‬
‭○‬ ‭Pooling Layers‬‭: Keep reducing the feature map dimensions.‬

‭6. Flattening Layer‬

‭●‬ P
‭ urpose‬‭: Converts the 2D feature maps into a‬‭1D vector‬‭to feed into the fully‬
‭connected layers.‬

‭7. Fully Connected Layer (Dense Layer)‬

‭●‬ P ‭ urpose‬‭: Connects all features learned from previous layers and combines them‬
‭to make a decision.‬
‭●‬ ‭Operation‬‭: Computes the weighted sum of the input features to classify the‬
‭image.‬

‭8. Output Layer‬

‭‬ P
● ‭ urpose‬‭: Provides the‬‭final prediction‬‭(defective or non-defective part).‬
‭●‬ ‭Activation Function‬‭:‬
‭○‬ ‭Sigmoid‬‭: For binary classification (e.g., Defective = 1, Non-Defective =‬
‭0).‬

‭14‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭○‬ S
‭ oftmax‬‭: For multi-class classification (e.g., crack, missing part, or‬
‭scratch).‬

‭6.‬ I‭n Bayesian learning, the prior and posterior probabilities play a key role. Explain‬
‭these concepts and apply them to determine the probability of tool wear given‬
‭observed cutting force and vibration data.‬

‭●‬ B ‭ ayesian Learning uses Bayes' theorem to make predictions and update‬
‭probabilities as new data becomes available.‬
‭●‬ ‭Bayes’ Theorem:‬
‭𝑃‬(‭𝐷‬‭|‬‭𝐻‬)‭‬×‭𝑃‬(‭𝐻‬)
‭𝑃‬(‭𝐻‬‭|‬‭𝐷‬)‭‬ = ‭‬ ‭𝑃‬(‭𝐷‬)

‭Where:‬
‭○‬ ‭𝑃(𝐻∣𝐷): Posterior probability (probability of hypothesis 𝐻 given data 𝐷).‬
‭○‬ ‭𝑃(𝐷∣𝐻): Likelihood (probability of data 𝐷 given hypothesis 𝐻).‬
‭○‬ ‭𝑃(𝐻): Prior probability (belief about 𝐻 before seeing data).‬
‭○‬ ‭𝑃(𝐷): Evidence (probability of data 𝐷).‬

‭Application: Determining Probability of Tool Wear‬

‭Let’s‬ ‭consider‬ ‭a‬ ‭tool‬ ‭wear‬ ‭detection‬ ‭problem‬ ‭in‬ ‭a‬ ‭manufacturing‬ ‭setting‬ ‭where‬ ‭cutting‬
t‭ools are monitored for wear. We observe cutting force and vibration data as indicators.‬

‭1.‬ ‭Hypothesis (H):‬

‭○‬ ‭H1: Tool is worn (tool wear exists).‬
‭○‬ ‭H2​: Tool is not worn.‬
‭2.‬ ‭Observed Data (D):‬
‭○‬ ‭Cutting force and vibration levels measured from the machine.‬

‭Steps to Apply Bayesian Learning‬

‭1.‬ ‭Prior Probability P(H)‬

‭ he‬ ‭prior‬ ‭probability‬ ‭reflects‬ ‭the‬ ‭initial‬ ‭belief‬ ‭about‬ ‭the‬ ‭likelihood‬ ‭of‬ ‭tool‬ ‭wear‬
T
‭based on past experience. For example:‬

‭●‬ P ‭ (H1)=0.3‬ ‭→‬ ‭From‬ ‭previous‬ ‭data,‬ ‭30%‬ ‭of‬ ‭tools‬ ‭tend‬‭to‬‭wear‬‭out‬‭during‬
‭similar operations.‬
‭●‬ ‭P(H2)=0.7 → 70% of tools remain in good condition.‬

‭2.‬ ‭Likelihood P(D∣H)‬

‭15‬
‭ I for ME‬
A
‭Department of Mechanical Engineering‬

‭ he‬‭likelihood‬‭is‬‭the‬‭probability‬‭of‬‭observing‬‭the‬‭cutting‬‭force‬‭and‬‭vibration‬‭levels‬
T
‭given the state of the tool. For example:‬

‭●‬ I‭f the tool is worn (H1​), the likelihood of high vibration and cutting force:‬
‭P(D∣H1)=0.8‬
‭●‬ ‭If the tool is not worn (H2​), the likelihood of high vibration and cutting‬
‭force: P(D∣H2)=0.2‬

‭3.‬ ‭Evidence P(D)‬

‭ he‬‭evidence‬‭is‬‭the‬‭total‬‭probability‬‭of‬‭observing‬‭the‬‭data‬‭DDD‬‭regardless‬‭of‬‭the‬
T
‭tool state. This is calculated as:‬

‭P(D) = P(D∣H1)⋅P(H1) + P(D∣H2)⋅P(H2)‬

‭Substitute the values:‬

‭P(D) = (0.8 X 0.3) + (0.2 X 0.7) = 0.24 + 0.14 = 0.38‬

‭4.‬ ‭Posterior Probability P(H∣D)‬

‭ he‬ ‭posterior‬ ‭probability‬ ‭updates‬ ‭the‬ ‭belief‬ ‭about‬ ‭tool‬‭wear‬‭given‬‭the‬‭observed‬

T
‭data. Using Bayes’ theorem:‬

‭𝑃‬(‭𝐷‬‭|‬‭𝐻‬‭1‬)‭‬×‭𝑃‬(‭𝐻‬‭1‬)
‭𝑃‬(‭𝐻‬‭1|‬‭𝐷‬)‭‬ = ‭‬ ‭𝑃‬(‭𝐷‬)

‭Substitute the values:‬

‭0‬.‭8‬‭‬×‭0‬.‭3‬
‭𝑃‬(‭𝐻‬‭|‬‭𝐷‬)‭‬ = ‭‬ ‭0‬.‭38‬
= ‭0‬. ‭63‬
‭Thus, the updated probability of tool wear is 63%.‬

‭ .‬ P
1 ‭ rior Belief: Initially, there was a 30% chance that the tool was worn.‬
‭2.‬ ‭Updated Belief (Posterior): After observing the high vibration and cutting‬
‭force, the probability of tool wear increased to 63%.‬

‭ his‬ ‭means‬ ‭that‬ ‭the‬ ‭observed‬ ‭data‬ ‭significantly‬ ‭increases‬ ‭confidence‬ ‭that‬ ‭the‬
T
‭tool is worn, guiding the maintenance decision to replace or inspect the tool.‬

‭16‬