What Is Sequential Diagnosis?

Sequential diagnosis is a methodology for identifying faults in a system by applying a series of tests in an adaptive order, where the selection of each subsequent test depends on the outcomes of all previous tests. Rather than applying a predetermined, fixed set of test patterns, a sequential diagnostic procedure dynamically prunes the set of candidate fault hypotheses as evidence accumulates and selects the next test to maximally discriminate among the remaining possibilities. The goal is to isolate the fault to a specific component or location in the minimum expected number of tests.

The field draws on information theory, combinatorial optimization, and digital circuit testing. Its theoretical roots connect to the broader problem of sequential statistical decision-making, but its practical applications are concentrated in electronic design automation, where manufacturing defects and in-field failures in digital circuits require efficient and accurate identification. Related approaches appear in fault-tolerant systems, medical diagnosis, and process monitoring, wherever the cost of each test is significant and reducing the number of tests needed has direct practical value.

Fault Models and Test Generation

Sequential diagnosis operates against a model of the faults a system might exhibit. For digital circuits, the stuck-at fault model assumes that a signal line is permanently fixed at logic 0 or logic 1 regardless of the driving logic. Bridging faults model unwanted electrical connections between two signal lines. More complex fault models address delay faults and transition faults relevant to high-speed circuits.

For each fault model, a test pattern is a specific combination of input values applied to the circuit whose output response distinguishes one fault hypothesis from another. The diagnostic test generation problem is to construct a sequence of such patterns that isolates the actual fault with high confidence. The IEEE paper on fault diagnosis in synchronous sequential circuits based on effect-cause analysis presents a generation method that backtracks from observed output discrepancies to identify the most likely fault sites, applicable to circuits where state memory complicates direct analysis.

Sequential Testing Strategy and Adaptive Algorithms

The adaptive nature of sequential diagnosis distinguishes it from non-adaptive test sets. In a non-adaptive approach, all tests are determined before any are applied, and the fault is identified by matching the full response vector against a fault dictionary. In sequential diagnosis, each test response partitions the remaining fault candidates into groups, and the next test is chosen to partition those groups as evenly as possible, minimizing the maximum or expected remaining uncertainty.

The IEEE paper on sequential testing algorithms for multiple fault diagnosis establishes the computational complexity of optimal multiple-fault isolation as super-exponential, motivating heuristic approaches that trade optimality for tractable computation. Diagnostic directed graphs, which encode the discrimination structure among fault candidates, offer a compact representation that supports efficient adaptive test selection without constructing an explicit diagnostic tree.

Built-in self-test (BIST) architectures implement diagnostic capability directly in hardware, allowing circuits to test and partially diagnose themselves during power-on or periodic maintenance windows. The IEEE paper on built-in self-test methods for diagnosis of synchronous sequential circuits describes BIST techniques that generate test sequences on-chip and analyze responses using hardware comparators, reducing the dependency on external test equipment.

Applications

Sequential diagnosis has applications in a wide range of fields, including:

• Manufacturing test of digital integrated circuits and system-on-chip devices
• In-field diagnosis of field-programmable gate arrays and embedded processors
• Fault isolation in aerospace and automotive electronics where repair access is constrained
• Medical diagnostic systems that apply tests in sequence to refine differential diagnoses
• Network fault localization in telecommunications infrastructure