Q: You are controlling a furnace at 850°C and observe PID integral windup - the temperature overshoots the setpoint by 15°C before coming down. How do you diagnose and fix this? When should you use auto-tune?

Answer in 3 parts: (1) Diagnosing integral windup - the integral (I-term) accumulates large error while the actuator is saturated (valve 100% open or heater at full power); when error reaches zero the I-term is still large so it continues to drive output excessively; check by plotting I-term and output in the SCADA historian - if output is saturated but I-term keeps growing = windup; (2) Fixing windup - Anti-windup techniques: clamping (stop integrating when output saturates), back-calculation (feed output error back into integrator), conditional integration (only integrate when error is within a small band). In Siemens PID_Compact: enable the 'Anti-windup' parameter; (3) When to auto-tune - use when no mathematical process model is available, after replacing sensor or actuator, when process characteristics change due to wear. Do not auto-tune when the process is unstable or production is running (auto-tune deliberately causes disturbance). Good follow-up: 'Would you choose PID or Cascade control for this furnace?' - Cascade (outer loop: temperature, inner loop: heater current) responds faster to disturbances. Senior candidates will mention Smith Predictor for processes with large dead-time.

Q: You are tasked with designing an OEE dashboard for an SMT (Surface Mount Technology) line at an electronics factory. The SCADA already has raw data. How would you design the dashboard so a production manager can make a decision in under 60 seconds?

Design following a 'decision-first' principle: (1) Tier 1 KPIs (top of page, large numbers) - current OEE % with RAG color coding (Red 85%), real-time comparison against target, 8-hour trend; (2) Breakdown into 3 OEE components - Availability (actual runtime / planned time, top 3 downtime reasons), Performance (actual speed / design speed, slow-run instances), Quality (good parts / total parts, top defect types with Pareto chart); (3) Drill-down - click Availability to view downtime history with timestamps from SCADA event log; (4) Shift comparison - today vs. yesterday vs. same shift last week; (5) Data sources from SCADA - event log for downtime calculation, counter I/O for part count, defect signal from AOI (Automated Optical Inspection) machine. Refresh rate: 15-60 seconds (millisecond real-time not needed for OEE). Follow up on data quality: 'How do you distinguish planned downtime (meals, meetings) from unplanned downtime (machine breakdown) in SCADA data?' Strong candidates describe using reason codes entered by operators on the HMI when stopping the machine, and a scheduled downtime calendar in the MES.

Question 1

Explain the 4 phases of the PLC scan cycle on a Siemens S7-1500. If your control loop requires a 100ms cycle, how would you optimize the program to ensure no deadline is missed?

Accepted Answer

Answer in 2 parts: (1) 4 scan cycle phases - Input Scan (reads I/O states into process image table), Program Execution (executes OB1 and FC/FB blocks), Output Scan (writes process image to physical I/O), Communication/Housekeeping (processes HMI requests, network communication). (2) Optimizing for 100ms loop - use cyclic interrupt OB (OB30 at 100ms interval) instead of OB1 for real-time control, move critical control logic into OB30, use Direct I/O Access (bypass process image) for time-critical signals, disable unnecessary FCs, monitor Cycle Time in TIA Portal diagnostics to detect overruns. Strong candidates distinguish OB1 (free-running) from cyclic interrupt OBs (OB30-OB38 with fixed intervals). Follow up: 'If OB30 is overrun, how does the S7-1500 handle it?' - the correct answer is OB80 (Time Error OB) is called to handle the error.

Question 2

According to IEC 61131-3, when do you choose Structured Text (ST) over Ladder Logic (LD) for PLC programming in the Vietnamese factory context? Give a concrete example for each language.

Accepted Answer

Analyze across 4 criteria: (1) Logic complexity - LD suits simple relay logic (interlocks, start/stop circuits), ST for complex algorithms (PID calculation, recipe management, string processing); (2) Code readers - LD is familiar to Vietnamese maintenance engineers with relay logic background, ST resembles C/Pascal and is easier for software engineers; (3) LD example: emergency stop interlock with 3 signal sources - visual, easy to trace faults; (4) ST example: calculating product weight/dimension ratios, sorting data arrays, processing recipe strings for food lines like Vinamilk. Best practice: use LD for I/O logic, call FBs written in ST for calculations. This question tests practical thinking, not just theory. At Samsung/Foxconn Vietnam, maintenance teams usually only understand Ladder - strong candidates will mention maintainability when choosing a language, not just personal preference.

Question 3

Compare Modbus RTU, OPC UA, and MQTT in the context of IIoT data collection at a Samsung Vietnam factory and connecting to AWS IoT Core. Which stack would you choose and why?

Accepted Answer

Analyze across 3 layers: (1) Field-to-Edge - Modbus RTU: legacy protocol (1979), master-slave, RS-485, only basic data types, no security, but ubiquitous in legacy equipment in VN; OPC UA: IEC 62541 standard, certificate-based security, rich data model, pub-sub from OPC UA 1.04, but heavier and more complex; MQTT: lightweight (lowest overhead), pub-sub, broker-based, ideal for edge-to-cloud, no industrial data model. (2) Recommended stack for Samsung/AWS: Modbus RTU from legacy devices -> Edge Gateway (local MQTT broker) -> AWS IoT Core (MQTT over TLS). OPC UA if new devices support it. (3) AWS IoT Core only supports MQTT/HTTPS, cannot speak Modbus RTU directly - requires Edge Gateway (AWS Greengrass or Moxa, Advantech). Follow up on security: 'Modbus RTU has no authentication - how do you protect the field-to-edge layer?' Strong candidates mention network segmentation (VLAN for OT), encrypted VPN tunnels, and AWS IoT Core X.509 certificates per device.

Question 4

Describe a SCADA redundancy architecture to ensure failover from primary to backup server in under 30 seconds without data loss. Have you implemented or designed such a solution?

Accepted Answer

Design across 4 components: (1) Hot Standby Architecture - primary and backup servers run simultaneously, backup receives the full data stream like primary (active-passive or active-active), heartbeat signal every 1-5 seconds to detect failure; (2) Data synchronization - distributed historian replication (OSIsoft PI Mirror, Wonderware Historian Replication) ensures no data points are lost during failover, all writes acknowledged when both nodes receive; (3) Failover trigger - missing 3 consecutive heartbeats (15s) + check database connectivity before promoting backup, total time < 30s; (4) Network redundancy - dual NICs on each server, redundant switches, no single point of failure at the network layer. Practical platforms: Wonderware System Platform, Ignition SCADA (Redundancy Module), WinCC Redundancy. Senior differentiator: knowing the difference between 'warm standby' (backup syncs periodically, possible data loss) and 'hot standby' (real-time sync, zero data loss). Ask about RPO (Recovery Point Objective) and RTO (Recovery Time Objective) - this is the language of business stakeholders.

Question 5

You are controlling a furnace at 850°C and observe PID integral windup - the temperature overshoots the setpoint by 15°C before coming down. How do you diagnose and fix this? When should you use auto-tune?

Accepted Answer

Answer in 3 parts: (1) Diagnosing integral windup - the integral (I-term) accumulates large error while the actuator is saturated (valve 100% open or heater at full power); when error reaches zero the I-term is still large so it continues to drive output excessively; check by plotting I-term and output in the SCADA historian - if output is saturated but I-term keeps growing = windup; (2) Fixing windup - Anti-windup techniques: clamping (stop integrating when output saturates), back-calculation (feed output error back into integrator), conditional integration (only integrate when error is within a small band). In Siemens PID_Compact: enable the 'Anti-windup' parameter; (3) When to auto-tune - use when no mathematical process model is available, after replacing sensor or actuator, when process characteristics change due to wear. Do not auto-tune when the process is unstable or production is running (auto-tune deliberately causes disturbance). Good follow-up: 'Would you choose PID or Cascade control for this furnace?' - Cascade (outer loop: temperature, inner loop: heater current) responds faster to disturbances. Senior candidates will mention Smith Predictor for processes with large dead-time.

Question 6

You are tasked with designing an OEE dashboard for an SMT (Surface Mount Technology) line at an electronics factory. The SCADA already has raw data. How would you design the dashboard so a production manager can make a decision in under 60 seconds?

Accepted Answer

Design following a 'decision-first' principle: (1) Tier 1 KPIs (top of page, large numbers) - current OEE % with RAG color coding (Red <60%, Amber 60-85%, Green >85%), real-time comparison against target, 8-hour trend; (2) Breakdown into 3 OEE components - Availability (actual runtime / planned time, top 3 downtime reasons), Performance (actual speed / design speed, slow-run instances), Quality (good parts / total parts, top defect types with Pareto chart); (3) Drill-down - click Availability to view downtime history with timestamps from SCADA event log; (4) Shift comparison - today vs. yesterday vs. same shift last week; (5) Data sources from SCADA - event log for downtime calculation, counter I/O for part count, defect signal from AOI (Automated Optical Inspection) machine. Refresh rate: 15-60 seconds (millisecond real-time not needed for OEE). Follow up on data quality: 'How do you distinguish planned downtime (meals, meetings) from unplanned downtime (machine breakdown) in SCADA data?' Strong candidates describe using reason codes entered by operators on the HMI when stopping the machine, and a scheduled downtime calendar in the MES.

Question 7

At Samsung or Foxconn in Vietnam, how is the LOTO (Lockout/Tagout) procedure performed before machine maintenance? Describe each step according to OSHA 29 CFR 1910.147.

Accepted Answer

8 LOTO steps per OSHA 29 CFR 1910.147: (1) Notify - inform all affected employees the equipment will be locked out; (2) Identify energy sources - electrical, pneumatic, hydraulic, thermal, gravitational (e.g., a raised actuator that has not been lowered); (3) Isolate - shut down the machine following the standard shutdown procedure, move all energy isolation devices to OFF/SAFE position; (4) Lockout/Tagout - each maintenance engineer applies their PERSONAL padlock to the energy isolation device plus a name tag; (5) Release residual energy - discharge capacitors, bleed pneumatic pressure, lower suspended components; (6) Verify isolation - try pressing the start button, use a voltmeter to confirm no voltage; (7) Perform maintenance; (8) Restore - remove LOTO in reverse order, notify employees, perform controlled restart. Key point: only the person who applied the lock may remove it. At large FDI companies like Samsung Thai Nguyen and Foxconn Bac Giang, LOTO is mandatory per parent corporation global standards - violations can result in immediate termination. Ask the candidate: 'If the previous shift applied a lock but has gone home, how do you handle it?' - correct answer: contact supervisor, never remove another person's lock, there is a documented emergency removal procedure.

Question 8

Foxconn Bac Giang is recognized as a WEF Lighthouse Factory. What is a Digital Twin and how is this technology applied at an electronics factory following the Foxconn model? If tasked with building a Digital Twin for an assembly line, where would you start?

Accepted Answer

Answer in 3 parts: (1) Digital Twin definition - a real-time digital replica of a physical asset, continuously synchronized via IoT sensors. Not a static simulation but a living model that updates with actual machine state; (2) Foxconn Bac Giang Lighthouse applications: Digital Twin of iPhone/MacBook assembly line to predict bottlenecks before they occur, simulation before physical line changes (reducing changeover time from 8h to 2h), predictive maintenance from vibration/temperature data of robot arms and conveyors, virtual commissioning (testing PLC programs on the digital twin before deploying to real machines); (3) Build roadmap - start with data foundation: (a) deploy sensors and connect SCADA/MES, (b) build 3D model of the line (import CAD into Unity or Siemens NX/Teamcenter, NVIDIA Omniverse), (c) bind real-time data to 3D model via API, (d) add analytics layer (prediction, optimization), (e) first use case should be small with clear ROI - e.g., predicting downtime for one machine. This question separates candidates who actually work in Industry 4.0 from those who only read about it. Follow up: 'What is the difference between a Digital Twin and a Digital Shadow?' - Digital Shadow is one-way data flow (physical -> digital, no feedback), Digital Twin is bidirectional (digital can control back to physical). Candidates who know Siemens Xcelerator, PTC ThingWorx, AVEVA, or NVIDIA Omniverse is a good sign.